Want the conclusions? Start with the findings in plain language. This page contains the individual tests, exact prompt changes and measurements.
100/100 pairs ready · seed 28510931 · 960×544 · 243 frames / 24 fps · Hyperflow 8-step · MotionCache OFF.
AI-generated research clips. Exploratory, one seed and mostly one shared scene/reference set. This is not a pass/fail certification of the guide. Frame/audio differences are not quality scores. Licensing and attribution.
Part 1 analysis · Blind scoring panel · Reference assets · Measurements
Highlights: repeat control · remove retention · lighting transfer · cut vs dissolve · partial audio copy. The analysis is unblinded; dialogue and most music judgments remain unassessed.
Start with the blind panel if you want unbiased arm labels. Below: A is left/top; B is right/bottom. Video streams are native, remuxed without re-encoding.
What changed: None: genuinely rerender the same graph and seed.
What we checked: Decoded-frame/audio differences; fresh sampler execution proof.
What we observed: The two fresh rerenders have identical native decoded frames and identical decoded PCM audio. Different MP4 hashes come from container/metadata differences. This is a clean repeat control for this run, not proof of determinism on every configuration.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
# Identical-prompt repeat control: no prompt difference.
Downscaled RGB MAE: 0.0/255. Audio RMS: A -27.757 dBFS; B -27.757 dBFS.
What changed: Remove entire retention_analysis block.
What we checked: Reference binding, scripted action and unintended speech/music.
What we observed: Both versions keep the two adults and perform the camera handoff. Without retention_analysis (B), the puppy is on the floor mat rather than the bench seen in A; B better follows that specific spatial instruction. The block is not necessary for the core handoff in this sample. Audio semantics unassessed.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -7,13 +7,6 @@ summary: [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. - -retention_analysis: -<Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. -<Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. -<Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 25.0661/255. Audio RMS: A -27.757 dBFS; B -26.472 dBFS.
What changed: Remove entire summary block.
What we checked: Reference binding, scripted action and unintended speech/music.
What we observed: Removing the summary still produces the handoff and recognizable cast. Table geometry and the card's orientation change; the summary is not necessary for basic scene survival in this sample.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -4,9 +4,6 @@ <Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. <Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. <Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. - -summary: -[reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions.
Downscaled RGB MAE: 25.7994/255. Audio RMS: A -27.757 dBFS; B -26.71 dBFS.
What changed: Remove entire subject_definitions block.
What we checked: Reference binding, scripted action and unintended speech/music.
What we observed: Removing subject_definitions still gives the referenced-looking adults, cafe, dog and camera handoff. The dog remains on furniture in both. This prompt retains descriptive appearance and other reference-related prose, so it is not evidence that reference binding never needs definitions.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,10 +1,3 @@ -subject_definitions: -<Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. -<Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. -<Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. -<Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. -<Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. - summary: [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches.
Downscaled RGB MAE: 24.4146/255. Audio RMS: A -27.757 dBFS; B -26.736 dBFS.
What changed: Remove entire overall_soundscape block.
What we checked: Reference binding, scripted action and unintended speech/music.
What we observed: Both rows preserve the cafe staging and handoff despite removing overall_soundscape. Audio cannot be judged from these frames; the actual soundscape ablation remains unscored.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -22,8 +22,5 @@ The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. -overall_soundscape: -Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. - non_diegetic_music: N/A
Downscaled RGB MAE: 15.1786/255. Audio RMS: A -27.757 dBFS; B -27.612 dBFS.
What changed: Remove entire non_diegetic_music block.
What we checked: Reference binding, scripted action and unintended speech/music.
What we observed: Removing the non_diegetic_music field leaves the visual handoff intact with modest framing changes. Its actual target, whether omitting N/A changes unwanted music, requires listening and is not assessed here.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -24,6 +24,3 @@ overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. - -non_diegetic_music: -N/A
Downscaled RGB MAE: 13.6865/255. Audio RMS: A -27.757 dBFS; B -25.361 dBFS.
What changed: Section order only.
What we checked: Adherence and reference binding.
What we observed: Reversing section order changes framing and staging details but retains the adults, dog, cafe and transfer. No obvious visual collapse from section order in this scene.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,12 +1,15 @@ -subject_definitions: -<Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. -<Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. -<Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. -<Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. -<Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. +non_diegetic_music: +N/A -summary: -[reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. +overall_soundscape: +Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. + +detailed_description: +A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. +[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. +The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. +At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. @@ -15,15 +18,12 @@ <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. -detailed_description: -A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. -From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +summary: +[reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. -overall_soundscape: -Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. - -non_diegetic_music: -N/A +subject_definitions: +<Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. +<Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. +<Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. +<Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. +<Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background.
Downscaled RGB MAE: 31.4014/255. Audio RMS: A -27.757 dBFS; B -26.156 dBFS.
What changed: Headings only; all content retained.
What we checked: Same targets as baseline.
What we observed: Plain-English headings still give recognizable scene content and a completed handoff. Fine timing/framing differ; the pictures do not show canonical headings being required for this action.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,29 +1,29 @@ -subject_definitions: +references: <Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. <Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. <Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. <Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. <Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. -summary: +overview: [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. -retention_analysis: +preservation: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. <Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. -detailed_description: +scene: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. -overall_soundscape: +sounds: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. -non_diegetic_music: +score: N/A
Downscaled RGB MAE: 19.3877/255. Audio RMS: A -27.757 dBFS; B -23.214 dBFS.
What changed: Remove headings only.
What we checked: Binding, actions, audio.
What we observed: Removing headings retains the two adults, dog and camera transfer. Both put the dog on seating rather than the requested floor mat. No demonstrated quality benefit from deleting headings, only tolerance in this example.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,29 +1,23 @@ -subject_definitions: <Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. <Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. <Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. <Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. <Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. -summary: [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. -retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. <Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. -detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. -overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. -non_diegetic_music: N/A
Downscaled RGB MAE: 22.9395/255. Audio RMS: A -27.757 dBFS; B -23.794 dBFS.
What changed: Main field name only.
What we checked: Visual and audio adherence.
What we observed: Changing the main field name preserves the handoff but changes table arrangement and adds tabletop clutter in B. The alternate field name is not a hard parsing failure here.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -15,7 +15,7 @@ <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. -detailed_description: +integrated_multimodal_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action.
Downscaled RGB MAE: 26.8089/255. Audio RMS: A -27.757 dBFS; B -27.925 dBFS.
What changed: Compress description while preserving target-event checklist.
What we checked: Score each invariant target, not verbosity.
What we observed: The concise body still completes a camera transfer, but B changes the table/staging, includes extra tabletop items and ends with both people facing the viewer. Compression changes the result; this is not a pure word-count effect with identical wording.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -16,9 +16,7 @@ <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: -A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. -From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. +[Shot 1] A realistic continuous view of the referenced cafe. The woman <Subject 1> in her black top stands left of the table; the bearded man <Subject 2> in his cardigan stands right. One black camera <Subject 4> passes from her hands to his at 2–4 seconds. He rotates it at 4–6 seconds. One golden puppy <Subject 5> stays on a blue mat beside the left table leg and turns its head at 6–8 seconds. The green lamps, cream walls, menu boards, conduits and CAFE sign remain recognizable. A white card on the table reads "OPEN AT NINE". He lowers the camera at 8–10 seconds and both smile. The strap hangs naturally; hands, feet and shadows stay coherent. End in a settled composition. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 37.1908/255. Audio RMS: A -27.757 dBFS; B -26.686 dBFS.
What changed: Append redundant target restatements.
What we checked: Adherence, repeated actions, truncation evidence.
What we observed: Redundant long restatement preserves the basic transfer in the sampled frames. No visible systematic gain is established; repeated-action/audio behavior and truncation are not resolved by this sparse frame survey.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -21,6 +21,7 @@ From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. The woman's top stays black, the man's cardigan stays dark, the camera remains a single black object, the puppy stays on the blue floor mat, and the cafe fixtures retain their positions. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 24.7012/255. Audio RMS: A -27.757 dBFS; B -27.509 dBFS.
What changed: Remove summary task prefix only.
What we checked: Same visual targets.
What we observed: Omitting the task prefix still produces the cafe handoff, with changes in viewpoint and table details. The prefix is not essential to this particular reference-generation outcome.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -6,7 +6,7 @@ <Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. summary: -[reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. +Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions.
Downscaled RGB MAE: 23.2132/255. Audio RMS: A -27.757 dBFS; B -26.837 dBFS.
What changed: Incorrect task prefix; reference wiring unchanged.
What we checked: Unexpected composition anchoring.
What we observed: Replacing the reference-generation prefix with keyframe-completion wording still produces a reference-based cafe handoff, with staging changes. The actual Ref2VA model and reference inputs never change, so this tests conflicting text, not switching the model's inference mode.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -6,7 +6,7 @@ <Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. summary: -[reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. +[keyframe completion] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions.
Downscaled RGB MAE: 18.43/255. Audio RMS: A -27.757 dBFS; B -26.889 dBFS.
What changed: Description language; dialogue verbatim.
What we checked: Same compact target checklist.
What we observed: English and French compact descriptions both produce the people, camera and cafe, but B becomes a close seated/frontal composition while A keeps a wider table view. Both put the dog on furniture. Scene survival does not imply equivalent detailed adherence or translated speech.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -16,7 +16,7 @@ <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: -[Shot 1] In the cafe <Subject 3>, the woman <Subject 1> stands left of the table and the man <Subject 2> stands right. She passes the single black camera <Subject 4> to him between two and four seconds. The golden puppy <Subject 5> stays on its blue floor mat. The filming camera moves slowly to the right. At eight seconds they smile and hold their positions until the end. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +[Shot 1] Dans le cafe <Subject 3>, la femme <Subject 1> se tient a gauche de la table et l'homme <Subject 2> a droite. Elle lui passe l'unique appareil photo noir <Subject 4> entre deux et quatre secondes. Le chiot dore <Subject 5> reste sur son tapis bleu au sol. La camera de tournage se deplace lentement vers la droite. A huit secondes, ils sourient et gardent leur position jusqu'a la fin. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 43.062/255. Audio RMS: A -26.131 dBFS; B -25.14 dBFS.
What changed: Remove marker words, retain prose.
What we checked: Identity, count and requested action.
What we observed: Omitting the retention marker words but retaining explanations still gives the camera transfer and recognizable subjects. Framing and the printed-card placement change. No general marker irrelevance follows from this one case.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -9,11 +9,11 @@ [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. -<Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. -<Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. +<Subject 1> (appears in [Shot 1]): Keep the woman's face, hair and black top while permitting the described actions. +<Subject 2> (appears in [Shot 1]): Keep the man's face, beard, hair and cardigan while permitting the described actions. +<Subject 3> (appears in [Shot 1]): Keep the cafe architecture and fixtures while permitting the described camera movement. +<Subject 4> (appears in [Shot 1]): Keep one black camera with the same body and short lens during the handoff. +<Subject 5> (appears in [Shot 1]): Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 25.4406/255. Audio RMS: A -27.757 dBFS; B -27.11 dBFS.
What changed: Remove retention explanations, retain markers.
What we checked: Identity, count and requested action.
What we observed: Markers without explanatory sentences still produce the main handoff; the dog is small and on the floor mat in B, versus a larger bench-seated dog in A. Explanations are redundant with body prose here and do not enforce the floor placement in A.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -9,11 +9,11 @@ [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. -<Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. -<Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. +<Subject 1> (appears in [Shot 1]): fully_preserved +<Subject 2> (appears in [Shot 1]): fully_preserved +<Subject 3> (appears in [Shot 1]): fully_preserved +<Subject 4> (appears in [Shot 1]): fully_preserved +<Subject 5> (appears in [Shot 1]): fully_preserved detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 30.8886/255. Audio RMS: A -27.757 dBFS; B -25.983 dBFS.
What changed: Retention line breaks only.
What we checked: Identity, count and requested action.
What we observed: Combining retention lines into a paragraph changes small composition details but retains the principal subjects and handoff. No hard line-format requirement is evident.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -9,11 +9,7 @@ [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. -<Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. -<Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. +<Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. <Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 20.4199/255. Audio RMS: A -27.757 dBFS; B -27.467 dBFS.
What changed: Shot occurrence annotations only.
What we checked: Identity, count and requested action.
What we observed: Removing shot-occurrence annotations leaves the single-shot handoff intact. This single-shot test cannot settle whether those annotations help selective appearance across multiple shots.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -9,11 +9,11 @@ [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. -<Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. -<Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. +<Subject 1>: fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. +<Subject 2>: fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. +<Subject 3>: fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. +<Subject 4>: fully_preserved - Keep one black camera with the same body and short lens during the handoff. +<Subject 5>: fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 24.6099/255. Audio RMS: A -27.757 dBFS; B -26.946 dBFS.
What changed: One retention marker only; prose unchanged.
What we checked: Yellow clothing with unchanged face.
What we observed: Both partially_preserved (A) and fully_preserved (B) produce the requested yellow jacket while retaining broadly similar face/hair and the handoff. The subject definition already permits replacing the top. Since the guide scopes preservation to the defined role, this is not a clean contradiction of its fully_preserved guidance.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -9,7 +9,7 @@ [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1]): partially_preserved - The woman wears a bright yellow rain jacket instead of her black top. Keep her face and hair unchanged. +<Subject 1> (appears in [Shot 1]): fully_preserved - The woman wears a bright yellow rain jacket instead of her black top. Keep her face and hair unchanged. <Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff.
Downscaled RGB MAE: 12.1007/255. Audio RMS: A -26.509 dBFS; B -26.583 dBFS.
What changed: One retention marker only; prose unchanged.
What we checked: Red camera body and one object.
What we observed: Both marker choices produce a reddish camera body with a dark lens during the transfer. The definition specifies shape and lens but not the original body color, so fully_preserved need not prohibit the edit within that narrowed role. Fine geometry and exact hue are not scored.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -12,7 +12,7 @@ <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. <Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1]): partially_preserved - The camera body is red instead of black; its grip shape, short black lens and proportions remain unchanged. +<Subject 4> (appears in [Shot 1]): fully_preserved - The camera body is red instead of black; its grip shape, short black lens and proportions remain unchanged. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description:
Downscaled RGB MAE: 15.9091/255. Audio RMS: A -27.598 dBFS; B -27.373 dBFS.
What changed: One retention marker only; prose unchanged.
What we checked: Blue light on man only, neutral woman.
What we observed: A shows conspicuous blue light on the man's face from 2.5 through 9.5 s; B stays much more neutral. The woman's face is not strongly blue in either row. This is a task-aligned visual difference favoring attribute_transfer over fully_preserved for this seed, not a general estimate of its benefit.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -9,7 +9,7 @@ [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1]): attribute_transfer - In addition to retaining the woman's identity, transfer her portrait's blue side-lighting to the man's face. Her own face is neutrally lit. Keep both identities unchanged. +<Subject 1> (appears in [Shot 1]): fully_preserved - In addition to retaining the woman's identity, transfer her portrait's blue side-lighting to the man's face. Her own face is neutrally lit. Keep both identities unchanged. <Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff.
Downscaled RGB MAE: 16.1792/255. Audio RMS: A -27.193 dBFS; B -27.206 dBFS.
What changed: One retention marker only; prose unchanged.
What we checked: New architecture; unchanged people, prop and puppy.
What we observed: Both weak_reference and fully_preserved create the pale arched room and retain the people, dog and handoff. The definition explicitly requests only the cafe's welcoming atmosphere, not its fixtures. Preserving that role is compatible with changing architecture; this is not proof that the marker is ignored.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -11,7 +11,7 @@ retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. <Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1]): weak_reference - Replace the cafe fixtures with white arches, pale stone and green plants. Use the cafe reference only for a welcoming atmosphere, not its architecture. +<Subject 3> (appears in [Shot 1]): fully_preserved - Replace the cafe fixtures with white arches, pale stone and green plants. Use the cafe reference only for a welcoming atmosphere, not its architecture. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement.
Downscaled RGB MAE: 11.0254/255. Audio RMS: A -28.475 dBFS; B -28.374 dBFS.
What changed: Markers only, identical moving scene.
What we checked: Whether motion still occurs without appearance loss.
What we observed: Both variants show changing viewpoint, handoff and body action. fully_preserved does not freeze the people or composition here. The guide does not define new target actions as a loss of fidelity to an identity reference, so movement is not itself a guide violation.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -9,11 +9,11 @@ [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. -<Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. -<Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. +<Subject 1> (appears in [Shot 1]): partially_preserved - Keep the woman's face, hair and black top while permitting the described actions. +<Subject 2> (appears in [Shot 1]): partially_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. +<Subject 3> (appears in [Shot 1]): partially_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. +<Subject 4> (appears in [Shot 1]): partially_preserved - Keep one black camera with the same body and short lens during the handoff. +<Subject 5> (appears in [Shot 1]): partially_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 19.4624/255. Audio RMS: A -27.757 dBFS; B -26.802 dBFS.
What changed: Consistent subject-label spelling only.
What we checked: Identity, voice and object assignment.
What we observed: Named labels retain the recognizable adults, dog and handoff. No visual binding collapse; voice assignment remains unassessed.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,26 +1,26 @@ subject_definitions: -<Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. -<Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. -<Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. -<Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. -<Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. +Mara is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. +Leon is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. +the cafe is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. +the camera is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. +the puppy is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. summary: [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. -<Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. -<Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. +Mara (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. +Leon (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. +the cafe (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. +the camera (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. +the puppy (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +[Shot 1] The cafe the cafe contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman Mara, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man Leon, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera the camera is a single compact black object with a short lens and a strap. The puppy the puppy sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman Mara (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man Leon (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 26.0558/255. Audio RMS: A -27.757 dBFS; B -27.752 dBFS.
What changed: Use image labels in place of subject labels consistently.
What we checked: Identity and pose freedom.
What we observed: Using picture labels directly still allows a handoff and changing pose. It does not visibly lock the portrait composition in this sample.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,26 +1,26 @@ subject_definitions: -<Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. -<Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. -<Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. -<Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. -<Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. +<Picture 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. +<Picture 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. +<Picture 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. +<Picture 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. +<Picture 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. summary: [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. -<Subject 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. -<Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. +<Picture 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. +<Picture 2> (appears in [Shot 1]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. +<Picture 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. +<Picture 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. +<Picture 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +[Shot 1] The cafe <Picture 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Picture 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Picture 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Picture 4> is a single compact black object with a short lens and a strap. The puppy <Picture 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Picture 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Picture 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 19.5735/255. Audio RMS: A -27.757 dBFS; B -26.438 dBFS.
What changed: Add separate source-only picture entries.
What we checked: Binding and scene clutter.
What we observed: Extra source-only picture definitions do not create additional people/dogs in the sampled views. The handoff survives; this is not proof that redundant definitions are beneficial.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -4,6 +4,11 @@ <Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. <Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. <Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. +<Picture 1> supplies the visual reference for <Subject 1>. +<Picture 2> supplies the visual reference for <Subject 2>. +<Picture 3> supplies the visual reference for <Subject 3>. +<Picture 4> supplies the visual reference for <Subject 4>. +<Picture 5> supplies the visual reference for <Subject 5>. summary: [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches.
Downscaled RGB MAE: 17.0822/255. Audio RMS: A -27.757 dBFS; B -27.343 dBFS.
What changed: Incorrect woman subject index in description.
What we checked: Speaker/identity swap; assay sensitivity.
What we observed: The deliberately wrong subject index does not cause an obvious visual identity swap: a woman remains left and a man right while the camera is passed. Descriptive prose can compensate, making this a weak sensitivity control. Speaker assignment was not assessed.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -17,10 +17,10 @@ detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 2>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 2> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 20.5027/255. Audio RMS: A -27.757 dBFS; B -27.647 dBFS.
What changed: Remove repeated appearance clauses from body only.
What we checked: Identity retention.
What we observed: Removing repeated appearance clauses still preserves the recognizable cast and transfer. The source definitions and visual assets are still present, so this is only a redundancy ablation.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -17,7 +17,7 @@ detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, stands on the left. The man <Subject 2>, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 20.5473/255. Audio RMS: A -27.757 dBFS; B -27.913 dBFS.
What changed: Keep first tag per subject, remove later tags.
What we checked: Longer-range identity and voice binding.
What we observed: Removing repeated Subject labels at the dialogue mentions leaves the visual cast and handoff intact. Speaker IDs remain in both arms, and the target is actual voice-to-person binding; that needs listening/lip-sync review, so no binding winner is assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 17.5717/255. Audio RMS: A -27.757 dBFS; B -27.277 dBFS.
What changed: Definition line order only.
What we checked: Whether ordinal position overrides explicit IDs.
What we observed: Reordering definitions while keeping IDs produces the same cast roles and handoff. The dog remains on seating, not the floor mat.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,9 +1,9 @@ subject_definitions: +<Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. +<Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. +<Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. +<Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. <Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. -<Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. -<Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. -<Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. -<Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. summary: [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches.
Downscaled RGB MAE: 17.27/255. Audio RMS: A -27.757 dBFS; B -27.792 dBFS.
What changed: Swap source-picture indices only.
What we checked: Reference confusion; positive evidence the test detects mapping effects.
What we observed: Swapping source-picture numbers does not visibly swap the two identities in these frames. This attempted positive control did not elicit the intended gross error; it cannot establish that explicit mapping is reliable or irrelevant.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,6 +1,6 @@ subject_definitions: <Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. -<Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. +<Subject 2> is the adult man in <Picture 1>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. <Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. <Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. <Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background.
Downscaled RGB MAE: 19.601/255. Audio RMS: A -27.757 dBFS; B -27.693 dBFS.
What changed: Style sentence position only.
What we checked: Visual style and unwanted text.
What we observed: Moving the style sentence inside the first shot leaves broadly similar naturalistic visuals and handoff. No clear visual advantage is established.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -16,8 +16,7 @@ <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: -A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +[Shot 1] A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 15.3082/255. Audio RMS: A -27.757 dBFS; B -26.932 dBFS.
What changed: Remove explicit second identity-source link; keep all images connected.
What we checked: Identity consistency and source blending.
What we observed: Both rows are closely matched seated cafe scenes with the same-looking pair, dog and camera transfer. Omitting the explicit second identity-source link causes only subtle visible differences here; both still receive all connected images.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,5 +1,5 @@ subject_definitions: -<Subject 1> is the adult woman in <Picture 1> and <Picture 6>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. +<Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. <Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. <Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. <Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects.
Downscaled RGB MAE: 3.1249/255. Audio RMS: A -26.622 dBFS; B -26.222 dBFS.
What changed: Remove later shot tags only.
What we checked: Cut count, timestamp error, spatial continuity.
What we observed: Both versions show a wide view, a close-up of the handoff, then a return to the wide view, even without later shot tags in B. Explicit cut prose remains and is sufficient to obtain the broad sequence here.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,8 +19,8 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. Until 00:04.000 the woman offers the camera. Her fingers curve around its grip and support the short lens while the man raises both hands to receive it. The strap hangs freely and the puppy watches without approaching. Keep the white card visible beyond their forearms, without replacing the printed words or turning it into subtitles. -[Shot 2] At 00:04.000, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. -[Shot 3] At 00:08.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. +At 00:04.000, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. +At 00:08.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. Keep every shot spatially continuous across the same table; do not move the subjects between cuts. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 23.7931/255. Audio RMS: A -28.361 dBFS; B -27.35 dBFS.
What changed: Remove explicit cut timestamps; preceding action timing retained.
What we checked: Cut count, timestamp error, spatial continuity.
What we observed: Both retain the wide/close/wide sequence. Removing cut-boundary times moves B's two abrupt frame changes to 4.042 and 6.833 s, versus 4.208 and 8.042 s in A. The second cut is substantially early relative to the original 8 s target; B's boundary frames confirm the shot change.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,8 +19,8 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. Until 00:04.000 the woman offers the camera. Her fingers curve around its grip and support the short lens while the man raises both hands to receive it. The strap hangs freely and the puppy watches without approaching. Keep the white card visible beyond their forearms, without replacing the printed words or turning it into subtitles. -[Shot 2] At 00:04.000, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. -[Shot 3] At 00:08.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. +[Shot 2] the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. +[Shot 3] the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. Keep every shot spatially continuous across the same table; do not move the subjects between cuts. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 39.242/255. Audio RMS: A -28.422 dBFS; B -27.892 dBFS.
What changed: Timestamp notation only.
What we checked: Cut count, timestamp error, spatial continuity.
What we observed: Both timestamp notations produce the intended broad shot sequence. B returns wide by 7.5 s; A is still close at that point. Equivalent wording need not mean identical temporal placement.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -18,9 +18,9 @@ detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. -Until 00:04.000 the woman offers the camera. Her fingers curve around its grip and support the short lens while the man raises both hands to receive it. The strap hangs freely and the puppy watches without approaching. Keep the white card visible beyond their forearms, without replacing the printed words or turning it into subtitles. -[Shot 2] At 00:04.000, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. -[Shot 3] At 00:08.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. +Until 4 seconds the woman offers the camera. Her fingers curve around its grip and support the short lens while the man raises both hands to receive it. The strap hangs freely and the puppy watches without approaching. Keep the white card visible beyond their forearms, without replacing the printed words or turning it into subtitles. +[Shot 2] At 4 seconds, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. +[Shot 3] At 8 seconds, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. Keep every shot spatially continuous across the same table; do not move the subjects between cuts. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 35.073/255. Audio RMS: A -28.422 dBFS; B -27.309 dBFS.
What changed: Add first-shot timestamp.
What we checked: Cut count, timestamp error, spatial continuity.
What we observed: Adding an explicit opening timestamp retains the same sampled wide/close/wide sequence. No visible need for that timestamp in this example.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -17,7 +17,7 @@ detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +[Shot 1] At 00:00.000, The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. Until 00:04.000 the woman offers the camera. Her fingers curve around its grip and support the short lens while the man raises both hands to receive it. The strap hangs freely and the puppy watches without approaching. Keep the white card visible beyond their forearms, without replacing the printed words or turning it into subtitles. [Shot 2] At 00:04.000, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. [Shot 3] At 00:08.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut.
Downscaled RGB MAE: 19.5259/255. Audio RMS: A -28.422 dBFS; B -28.423 dBFS.
What changed: Contradict chronological cut time.
What we checked: Cut count, timestamp error, spatial continuity.
What we observed: Even the contradictory decreasing cut times still produce a plausible wide/close/wide sequence, with a changed return time. Survival of contradictory instructions is not evidence of precise timestamp compliance.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. Until 00:04.000 the woman offers the camera. Her fingers curve around its grip and support the short lens while the man raises both hands to receive it. The strap hangs freely and the puppy watches without approaching. Keep the white card visible beyond their forearms, without replacing the printed words or turning it into subtitles. [Shot 2] At 00:04.000, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. -[Shot 3] At 00:08.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. +[Shot 3] At 00:02.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. Keep every shot spatially continuous across the same table; do not move the subjects between cuts. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 41.3618/255. Audio RMS: A -28.422 dBFS; B -29.511 dBFS.
What changed: Renumber shot IDs consistently.
What we checked: Cut count, timestamp error, spatial continuity.
What we observed: Nonsequential shot IDs still produce the three broad views in the expected narrative order. Literal sequential numbering is not essential to this sample.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -9,18 +9,18 @@ [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches. retention_analysis: -<Subject 1> (appears in [Shot 1], [Shot 2], [Shot 3]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. -<Subject 2> (appears in [Shot 1], [Shot 2], [Shot 3]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. -<Subject 3> (appears in [Shot 1], [Shot 2], [Shot 3]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. -<Subject 4> (appears in [Shot 1], [Shot 2], [Shot 3]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. -<Subject 5> (appears in [Shot 1], [Shot 2], [Shot 3]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. +<Subject 1> (appears in [Shot 1], [Shot 7], [Shot 9]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions. +<Subject 2> (appears in [Shot 1], [Shot 7], [Shot 9]): fully_preserved - Keep the man's face, beard, hair and cardigan while permitting the described actions. +<Subject 3> (appears in [Shot 1], [Shot 7], [Shot 9]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. +<Subject 4> (appears in [Shot 1], [Shot 7], [Shot 9]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. +<Subject 5> (appears in [Shot 1], [Shot 7], [Shot 9]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. Until 00:04.000 the woman offers the camera. Her fingers curve around its grip and support the short lens while the man raises both hands to receive it. The strap hangs freely and the puppy watches without approaching. Keep the white card visible beyond their forearms, without replacing the printed words or turning it into subtitles. -[Shot 2] At 00:04.000, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. -[Shot 3] At 00:08.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. +[Shot 7] At 00:04.000, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. +[Shot 9] At 00:08.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. Keep every shot spatially continuous across the same table; do not move the subjects between cuts. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 17.5395/255. Audio RMS: A -28.422 dBFS; B -28.741 dBFS.
What changed: Transition type instruction.
What we checked: Hard cut versus overlap transition.
What we observed: Changing cuts to cross-dissolves produces visible blended wide/close views in B around 4.125-4.5 s, instead of A's abrupt change at 4.208 s. Both retain the later hard cut. This is a clear task-aligned wording response in this seed.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. Until 00:04.000 the woman offers the camera. Her fingers curve around its grip and support the short lens while the man raises both hands to receive it. The strap hangs freely and the puppy watches without approaching. Keep the white card visible beyond their forearms, without replacing the printed words or turning it into subtitles. -[Shot 2] At 00:04.000, the camera cuts to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. +[Shot 2] At 00:04.000, the camera cross-dissolves to a close view of their hands as the man receives it. The angle stays on the same side of the table, preserving screen direction. For a brief moment both adults support the same object. Then her hands open and withdraw while his fingers close around the grip. The short lens remains attached and the strap swings once beneath his wrist. The table edge and green lamp glow establish the same location in the background. [Shot 3] At 00:08.000, the camera cuts back to the original wide view as both smile and the puppy remains on its mat. He holds the received camera at waist height. Her empty hands rest naturally by her sides, their feet stay planted and the room fixtures do not migrate. Hold this composition to the end without an extra cut. Keep every shot spatially continuous across the same table; do not move the subjects between cuts. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 17.0449/255. Audio RMS: A -28.422 dBFS; B -28.0 dBFS.
What changed: Add one shot tag but explicitly retain continuous action.
What we checked: Unwanted cut or geography reset.
What we observed: An extra shot tag does not create an obvious new composition in B's sampled views. Across all adjacent frames, the maximum downscaled change remains about 2.49/255, versus 2.45 in A; no hard-cut-sized spike appears. This does not grade fine motion quality.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -18,7 +18,7 @@ detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. -From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. +From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. [Shot 2] At 00:04.000, during the same continuous camera move, from 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 21.07/255. Audio RMS: A -26.946 dBFS; B -29.219 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Both phrasings move toward a tighter view, but the technical zoom wording in A tightens the framing more strongly by the end. This is a candidate difference in magnitude, not proof that only one form controls zoom correctly.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera zooms in slowly from a medium-wide view to the man's face; the camera body stays still. +The camera stays in place while its focal length increases slowly from a medium-wide view to the man's face. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 36.4747/255. Audio RMS: A -26.306 dBFS; B -27.791 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Both technical and plain-language push-in instructions show progressively tighter staging. The sparse frames do not reliably separate lens zoom from physical dolly/parallax.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera pushes in slowly by half a meter toward the camera handoff. +The filming camera moves slowly forward by half a meter toward the camera handoff. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 22.9731/255. Audio RMS: A -27.75 dBFS; B -26.477 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Both pan phrasings change viewpoint around the handoff; no clear winner for the specified motion from this sparse frame comparison.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera pans right slowly through twenty degrees from the woman toward the man. +The filming camera stays in place and pivots its lens slowly twenty degrees to the right from the woman toward the man. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 22.9303/255. Audio RMS: A -27.602 dBFS; B -27.598 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Technical truck and matched plain-language translation yield closely similar staging and transfer. No strong visual advantage for the technical term is apparent.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera trucks right slowly by half a meter while staying level. +The filming camera moves sideways to the right slowly by half a meter while staying level. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 11.9033/255. Audio RMS: A -27.435 dBFS; B -27.213 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: The tilt phrasings give visibly different framing over time, but both keep the handoff. A precise angular camera-motion judgment is not supported by these five frames.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera tilts down slowly from the adults' faces toward their hands. +The filming camera stays in place and pivots its lens slowly downward from the adults' faces toward their hands. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 27.7233/255. Audio RMS: A -27.995 dBFS; B -27.848 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Technical pedestal wording and its plain-English equivalent preserve similar action and framing. These views do not demonstrate a calibrated half-meter downward camera translation; no technical-wording winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera pedestals down slowly by half a meter, keeping its lens level. +The whole filming camera moves down slowly by half a meter, keeping its lens level. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 11.7279/255. Audio RMS: A -27.401 dBFS; B -28.742 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Both arc phrasings change the view around the people and table while preserving their roles. No clear retention winner is apparent.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera makes a slow twenty-degree arc around the table, keeping the handoff centered. +The filming camera travels slowly around the table along a twenty-degree circular path, keeping the handoff centered. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 27.8265/255. Audio RMS: A -27.026 dBFS; B -27.797 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Both roll phrasings tilt the image orientation by the end while retaining the cast and handoff. Rotation is visible under both descriptions; exact amplitude is not scored.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera rolls clockwise slowly by ten degrees, then holds. +The filming camera rotates slowly ten degrees clockwise around its lens axis, then holds. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 23.5263/255. Audio RMS: A -26.112 dBFS; B -25.812 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: The two static-camera descriptions keep generally stable views while the people act. Fine drift requires full playback or measured background tracking; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera holds a static shot throughout. +The filming camera does not move or change focal length throughout. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 23.709/255. Audio RMS: A -28.285 dBFS; B -27.577 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Both shake descriptions keep the intended scene and handoff. Camera shake is a high-frequency temporal property and cannot be ranked from five stills.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The handheld camera shakes slightly throughout. +The handheld view has small irregular tremors throughout. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 16.8626/255. Audio RMS: A -27.189 dBFS; B -26.616 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Both tracking descriptions retain the handoff with small view changes. The scene gives little subject travel, so this is a weak test of sustained tracking.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera tracks the woman as she takes one step toward the table. +The filming camera follows the woman's one-step movement toward the table, keeping her the same size in frame. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 21.9159/255. Audio RMS: A -26.46 dBFS; B -28.789 dBFS.
What changed: Technical term versus matched plain-language motion.
What we checked: Motion direction, parallax, endpoint, unintended cuts.
What we observed: Both the technical POV instruction and its plain-English equivalent fail the requested man's-eye viewpoint: the man's face and body remain visible across the table. This is a shared failure, not a technical-vocabulary win.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2> stands on the right; his face is outside his own point of view, while his hands and dark sleeves can enter the lower view. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -A POV shot from the man's eyes shows the woman's approaching hands and camera. +The view is exactly what the man sees through his eyes: the woman's approaching hands and camera. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 15.9919/255. Audio RMS: A -26.351 dBFS; B -25.453 dBFS.
What changed: Small versus large amplitude.
What we checked: Visible requested camera-motion response.
What we observed: Small versus large amplitude produces different framing but no strikingly larger travel is apparent in B's five sampled views. Do not infer calibrated amplitude control from this example.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera trucks right with small amplitude at slow speed. +The camera trucks right with large amplitude at slow speed. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 20.0067/255. Audio RMS: A -27.606 dBFS; B -26.951 dBFS.
What changed: Slow versus fast.
What we checked: Visible requested camera-motion response.
What we observed: Slow versus fast wording retains closely related cafe staging. Camera speed is not quantified here, and the sparse frame survey does not establish reliable speed control.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The camera trucks right with small amplitude at slow speed. +The camera trucks right with small amplitude at fast speed. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 20.1535/255. Audio RMS: A -27.606 dBFS; B -26.796 dBFS.
What changed: Natural sentence versus labeled fields.
What we checked: Visible requested camera-motion response.
What we observed: Labeled camera fields and a natural sentence both preserve the scene and handoff. No obvious need for the labeled form in this example.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -19,7 +19,7 @@ A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. -The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. +Camera: truck right; amplitude: small; speed: slow. Reveal a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape:
Downscaled RGB MAE: 18.7783/255. Audio RMS: A -27.757 dBFS; B -27.247 dBFS.
What changed: Remove dialogue delimiters only.
What we checked: Exact words; spoken metadata; speaker.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says [English] Hold it gently. At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies [English] I have it. Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 23.9866/255. Audio RMS: A -27.757 dBFS; B -27.09 dBFS.
What changed: Remove language tags only.
What we checked: Language and exact words.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 20.5165/255. Audio RMS: A -27.757 dBFS; B -28.468 dBFS.
What changed: Remove speaker IDs only.
What we checked: Correct speaker and stable voice.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1>, in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2>, in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 21.8401/255. Audio RMS: A -27.757 dBFS; B -26.997 dBFS.
What changed: Reuse S1 for a different speaker.
What we checked: Voice collapse or speaker swap.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S1), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 13.4511/255. Audio RMS: A -27.757 dBFS; B -27.562 dBFS.
What changed: Consistent nonsequential speaker IDs.
What we checked: Voice binding versus order convention.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1> (S7), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S9), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 16.3376/255. Audio RMS: A -27.757 dBFS; B -26.792 dBFS.
What changed: Move delivery phrase inside spoken block.
What we checked: Unwanted spoken stage direction.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1> (S1), says <d>[English] In a calm mid-pitched voice, hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 18.7954/255. Audio RMS: A -27.757 dBFS; B -23.463 dBFS.
What changed: Remove final punctuation inside dialogue.
What we checked: Delivery and stopping.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 19.5183/255. Audio RMS: A -27.757 dBFS; B -25.148 dBFS.
What changed: Decorative punctuation only.
What we checked: Prosody and literal punctuation vocalization.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently~~~!!!</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 20.8059/255. Audio RMS: A -27.757 dBFS; B -26.179 dBFS.
What changed: Canonical voiceover phrase versus synonym.
What we checked: Audible narration without visible speech.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:02.000 the woman <Subject 1> (S1) says in an off-screen voiceover <d>[English] This camera means a lot to me.</d> Her visible lips remain completely closed. +At 00:02.000 the woman <Subject 1> (S1) narrates off screen <d>[English] This camera means a lot to me.</d> Her visible lips remain completely closed. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 19.5613/255. Audio RMS: A -26.834 dBFS; B -27.318 dBFS.
What changed: Remove explicit closed-lips clause.
What we checked: Unwanted lip movement.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:02.000 the woman <Subject 1> (S1) says in an off-screen voiceover <d>[English] This camera means a lot to me.</d> Her visible lips remain completely closed. +At 00:02.000 the woman <Subject 1> (S1) says in an off-screen voiceover <d>[English] This camera means a lot to me.</d> overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 18.2859/255. Audio RMS: A -26.834 dBFS; B -26.663 dBFS.
What changed: Compound versus separate ID spelling.
What we checked: Simultaneous intelligible voices.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:03.000 the woman <Subject 1> and man <Subject 2> (S1,S2) say together <d>[English] Ready for the picture.</d> Their voices overlap in unison. +At 00:03.000 the woman <Subject 1> and man <Subject 2> (S1) and (S2) say together <d>[English] Ready for the picture.</d> Their voices overlap in unison. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 27.3942/255. Audio RMS: A -24.417 dBFS; B -24.706 dBFS.
What changed: Remove cross-cut markers; continuity prose retained.
What we checked: No restart, gap or voice swap at cut.
What we observed: The sampled scene stays wide, and the largest adjacent-frame change is only about 2.65/255 in A and 2.61/255 in B after downscaling. No clear requested cut was established, so this is a weak test of speech continuity across an actual cut. Speech itself is unassessed.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,8 +20,8 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The view stays still within each shot; the only cut is the specified cut at four seconds. -At 00:03.000 the woman <Subject 1> (S1) says <d>[English] Please keep the <scenetrans></d> -[Shot 2] At 00:04.000, the camera cuts to the man while her same line continues uninterrupted: <Subject 1> (S1), off-screen, <d>[English] <scenetrans>strap around your wrist.</d> +At 00:03.000 the woman <Subject 1> (S1) says <d>[English] Please keep the </d> +[Shot 2] At 00:04.000, the camera cuts to the man while her same line continues uninterrupted: <Subject 1> (S1), off-screen, <d>[English] strap around your wrist.</d> overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 19.1123/255. Audio RMS: A -26.5 dBFS; B -25.287 dBFS.
What changed: Remove cutoff marker; truncation prose retained.
What we checked: Native ending cuts speech, no rushed completion.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:08.500 the woman <Subject 1> (S1) starts speaking at an ordinary pace: <d>[English] Before you take another picture, remember to<cutoff></d> The utterance is abruptly interrupted by the video ending, not completed early. +At 00:08.500 the woman <Subject 1> (S1) starts speaking at an ordinary pace: <d>[English] Before you take another picture, remember to</d> The utterance is abruptly interrupted by the video ending, not completed early. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 18.3227/255. Audio RMS: A -23.648 dBFS; B -22.882 dBFS.
What changed: Language tags only, French words retained.
What we checked: Language, exact words and accent.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[French] Tiens-le doucement.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[French] Je le tiens.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Tiens-le doucement.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] Je le tiens.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 15.2559/255. Audio RMS: A -30.348 dBFS; B -30.738 dBFS.
What changed: Duplicate spoken words in soundscape.
What we checked: Extra repeats, voice timing.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -23,7 +23,7 @@ At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: -Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. +Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. The woman says Hold it gently. The man says I have it. non_diegetic_music: N/A
Downscaled RGB MAE: 22.5402/255. Audio RMS: A -27.757 dBFS; B -27.2 dBFS.
What changed: Add unused speaker ID to silent puppy.
What we checked: Unwanted vocalization or speaker confusion.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -17,7 +17,7 @@ detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> (S3) sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 21.9663/255. Audio RMS: A -27.757 dBFS; B -27.488 dBFS.
What changed: Delivery verb only; lyric words preserved.
What we checked: Melody versus spoken delivery.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -20,7 +20,7 @@ [Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. -At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, sings a short rising melody <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, sings a short falling answer <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard.
Downscaled RGB MAE: 23.3209/255. Audio RMS: A -27.757 dBFS; B -26.981 dBFS.
What changed: Quotation syntax only.
What we checked: OCR exactness and unintended subtitles.
What we observed: Both native 5 s frames contain a printed card on the table, with different lettering/presentation. The small foreshortened letters are not a reliable exact OCR target at this resolution. Quotation-mark removal does not eliminate the card, but exact text accuracy and full-clip subtitle behavior remain ungraded.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -17,7 +17,7 @@ detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "OPEN AT NINE". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text OPEN AT NINE. The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 16.3308/255. Audio RMS: A -27.757 dBFS; B -27.737 dBFS.
What changed: Translate requested visible text.
What we checked: OCR exactness and unintended subtitles.
What we observed: Native 5 s frames show different card presentation, including an upright card in B, but the small/foreshortened lettering does not allow a confident exact-spelling and accent grade for both arms. No language/spelling winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -17,7 +17,7 @@ detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion. -[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "CAFÉ OUVERT". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. +[Shot 1] The cafe <Subject 3> contains a waist-high wooden table beneath green hanging lamps. The dark menu boards and CAFE sign remain attached to the rear cream wall; exposed conduits run above them. The woman <Subject 1>, recognizable by her swept-back dark hair, facial features and black top, stands on the left. The man <Subject 2>, recognizable by his short beard, brown hair and dark cardigan, stands on the right. Their faces remain different from one another. The camera <Subject 4> is a single compact black object with a short lens and a strap. The puppy <Subject 5> sits on a blue floor mat beside the left table leg, its golden fur and floppy ears clearly visible. Both adults wear dark trousers and stand with their feet on the floor. The table is empty except for a small white card bearing the printed text "CAFE OPEN". The camera is in the woman's hands, not on the table. Their left and right positions do not exchange as the view changes. From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line.
Downscaled RGB MAE: 16.2824/255. Audio RMS: A -26.427 dBFS; B -26.399 dBFS.
What changed: Line breaks only.
What we checked: Ambience and event synchronization.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -23,7 +23,8 @@ At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. overall_soundscape: -Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. +Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. +No other intelligible speech is heard. non_diegetic_music: N/A
Downscaled RGB MAE: 14.3943/255. Audio RMS: A -27.757 dBFS; B -27.672 dBFS.
What changed: Silence field value.
What we checked: PCM RMS/peak and listening for any sound.
What we observed: Both silence variants produce very low-level but nonzero audio: A RMS -52.559 dBFS, B -51.582 dBFS. Peak levels are -35.726 and -34.505 dBFS. Listening has not established whether the residual is perceptually significant; this is not digital silence.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -24,7 +24,7 @@ The entire video is completely silent. overall_soundscape: -N/A +Complete silence throughout. non_diegetic_music: N/A
Downscaled RGB MAE: 19.6478/255. Audio RMS: A -52.559 dBFS; B -51.582 dBFS.
What changed: Music field value only.
What we checked: Unrequested score.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -26,4 +26,4 @@ Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. non_diegetic_music: -N/A +No background music.
Downscaled RGB MAE: 17.3703/255. Audio RMS: A -27.757 dBFS; B -28.324 dBFS.
What changed: Move identical score request into body.
What we checked: Score present, no visible source or character reaction.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -21,9 +21,7 @@ From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. +Audience-only background music: A quiet plucked acoustic guitar plays evenly spaced notes at a slow tempo, fades in during the opening second and fades out during the last second. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. - -non_diegetic_music: -A quiet plucked acoustic guitar plays evenly spaced notes at a slow tempo, fades in during the opening second and fades out during the last second.
Downscaled RGB MAE: 18.4308/255. Audio RMS: A -29.978 dBFS; B -27.177 dBFS.
What changed: Instrument/tempo specification versus mood-only request.
What we checked: Requested instrumentation and tempo; do not equate different music with failure.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -26,4 +26,4 @@ Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. non_diegetic_music: -A quiet plucked acoustic guitar plays evenly spaced notes at a slow tempo, fades in during the opening second and fades out during the last second. +A warm, comforting and nostalgic score.
Downscaled RGB MAE: 18.9748/255. Audio RMS: A -29.978 dBFS; B -28.874 dBFS.
What changed: Move radio instruction to audience-only field.
What we checked: Visible source, reaction and audio source binding.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -21,10 +21,9 @@ From the opening until 00:02.000, she holds the camera at waist height with one hand around the grip and the other beneath the lens. He looks at the camera and opens his hands. Between 00:02.000 and 00:04.000, she passes it across the table. Both briefly support its weight; he receives the grip, then she withdraws her empty hands. From 00:04.000 to 00:06.000, he rotates it a quarter-turn toward himself while she watches. The strap hangs below their hands and swings briefly before settling. Between 00:06.000 and 00:08.000, the puppy turns its head toward the camera without leaving the mat. Between 00:08.000 and 00:10.000, the man lowers the camera to his waist and both smile. Hold that settled composition through the last fraction of a second. Fingers remain distinct around the grip; the lens never changes size and no second camera appears. Shadows remain attached to feet, paws and furniture. Their eyelines follow the object or partner named in the current action. The filming camera trucks right with small amplitude at slow speed, revealing a little more of the table's right edge without cutting. At 00:01.000 the woman <Subject 1> (S1), in a calm mid-pitched voice, says <d>[English] Hold it gently.</d> At 00:05.000 the man <Subject 2> (S2), in a low relaxed voice, replies <d>[English] I have it.</d> Each closes their lips after their own line. -A small radio on the table emits a quiet solo-guitar tune; the woman hears it and briefly turns toward it. overall_soundscape: Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. non_diegetic_music: -N/A +A small radio on the table emits a quiet solo-guitar tune; the woman hears it and briefly turns toward it.
Downscaled RGB MAE: 21.1146/255. Audio RMS: A -28.252 dBFS; B -28.144 dBFS.
What changed: Add emotional intent; concrete music instruction unchanged.
What we checked: Instrument/tempo adherence versus subjective preference.
What we observed: The target here requires listening, transcription and/or lip-sync review. It is not assessed in this visual/signal pass; no winner assigned.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -26,4 +26,4 @@ Quiet cafe room tone, clothing rustle and a soft camera-strap tap accompany the action. No other intelligible speech is heard. non_diegetic_music: -Quiet plucked guitar notes repeat at a slow tempo without percussion. +Quiet plucked guitar notes repeat at a slow tempo without percussion. This score should make the audience feel warm nostalgia and emphasize the tenderness of the exchange.
Downscaled RGB MAE: 19.9047/255. Audio RMS: A -29.695 dBFS; B -30.068 dBFS.
What changed: Task prefix only; video input and prose identical.
What we checked: Continuity from source ending versus replay.
What we observed: Both arms replay the source-like handoff: the woman initially holds the camera, then gives it to the man. This conflicts with the explicit request to continue after the source ends without repeating the handoff. The source storyboard confirms the same broad action order. It is a shared failure under this preset, not proof about all continuation models or prefixes.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -7,7 +7,7 @@ <Video 1> is the supplied moving cafe clip, used in the relationship stated below. summary: -[video continuation] Continue the scene from the last moment of <Video 1>. The man keeps the camera and moves it closer to his chest; do not restart the handoff. +[reference generation] Continue the scene from the last moment of <Video 1>. The man keeps the camera and moves it closer to his chest; do not restart the handoff. retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions.
Downscaled RGB MAE: 0.9378/255. Audio RMS: A -55.818 dBFS; B -55.986 dBFS.
What changed: Task prefix only; video input and prose identical.
What we checked: Source trajectory retained and targeted recolor.
What we observed: Both outputs closely follow the source staging and show a red/reddish camera body despite differing task prefixes. The requested prop edit is visible in both; no broad prefix advantage is established.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -7,7 +7,7 @@ <Video 1> is the supplied moving cafe clip, used in the relationship stated below. summary: -[video editing] Edit <Video 1> while retaining its action order and viewpoint. Change only the camera body to red. +[reference generation] Edit <Video 1> while retaining its action order and viewpoint. Change only the camera body to red. retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions.
Downscaled RGB MAE: 0.9227/255. Audio RMS: A -53.944 dBFS; B -56.221 dBFS.
What changed: Task prefix only; video input and prose identical.
What we checked: Motion transferred without unintended continuation.
What we observed: Both outputs closely follow the source-like handoff and view trajectory. The target itself also requests a handoff, so this is a weak discriminator of motion-only versus full-content reference use. Do not count source resemblance alone as proof of failure or success.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -7,7 +7,7 @@ <Video 1> is the supplied moving cafe clip, used in the relationship stated below. summary: -[reference generation] Use only the camera trajectory of <Video 1>, not its source action timeline. The woman starts with the camera again and the man watches. +[video continuation] Use only the camera trajectory of <Video 1>, not its source action timeline. The woman starts with the camera again and the man watches. retention_analysis: <Subject 1> (appears in [Shot 1]): fully_preserved - Keep the woman's face, hair and black top while permitting the described actions.
Downscaled RGB MAE: 0.9619/255. Audio RMS: A -50.926 dBFS; B -51.592 dBFS.
What changed: Audio relationship marker only.
What we checked: Sample alignment/correlation, tone event times, unwanted sounds.
What we observed: Both marker variants reproduce the source's five tone events at approximately 330/440/550/660/770 Hz in the 1/3/5/7/9-second windows. A/B zero-lag PCM correlation is 0.99897, but neither decoded waveform equals the source PCM. Preservation here is not bit-perfect copying; VAE/codec effects are included.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -15,7 +15,7 @@ <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. -<Audio 1>: fully_copy - Use <Audio 1> unchanged as the complete output soundtrack. Add no speech, effects or music. +<Audio 1>: reference - Use <Audio 1> unchanged as the complete output soundtrack. Add no speech, effects or music. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 13.2416/255. Audio RMS: A -31.07 dBFS; B -30.995 dBFS.
What changed: Audio relationship marker only.
What we checked: Sample alignment/correlation, tone event times, unwanted sounds.
What we observed: Both partially_copy and fully_copy retain all five source-tone events, including the forbidden 1-, 7- and 9-second events. Outside the requested 2-6 s window, RMS remains about -31 dBFS in both arms. Neither obeys the partial-copy/silence request in this sample.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -15,7 +15,7 @@ <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. -<Audio 1>: partially_copy - Use only seconds 2–6 of <Audio 1> at output seconds 2–6. All remaining time is silent. +<Audio 1>: fully_copy - Use only seconds 2–6 of <Audio 1> at output seconds 2–6. All remaining time is silent. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 17.6062/255. Audio RMS: A -31.223 dBFS; B -31.192 dBFS.
What changed: Audio relationship marker only.
What we checked: Sample alignment/correlation, tone event times, unwanted sounds.
What we observed: Both weak_reference and fully_copy retain approximately the original ascending tone frequencies and event windows. Neither shows the requested new lower-pitched pattern in this signal check. This does not establish subjective timbre equivalence.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -15,7 +15,7 @@ <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. -<Audio 1>: weak_reference - Use the slow pulse character of <Audio 1>, but generate a new lower-pitched bell pattern, not its exact signal. +<Audio 1>: fully_copy - Use the slow pulse character of <Audio 1>, but generate a new lower-pitched bell pattern, not its exact signal. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 8.8247/255. Audio RMS: A -31.276 dBFS; B -31.142 dBFS.
What changed: Audio marker only; request new words in both.
What we checked: Voice similarity and leakage of source words.
What we observed: Voice identity, new-word correctness and leakage of the source sentence require listening and/or transcription. These criteria are not assessed by the visual/signal pass.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -15,7 +15,7 @@ <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. -<Audio 1>: reference - use the voice texture for new words, without copying the source sentence. +<Audio 1>: fully_copy - use the voice texture for new words, without copying the source sentence. detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 16.864/255. Audio RMS: A -28.385 dBFS; B -28.398 dBFS.
What changed: Remove S1 only from voice-reference definition.
What we checked: Voice-to-person binding.
What we observed: Voice-to-person binding is not assessed. Similar images or waveform differences cannot establish which character speaks in the reference voice.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -4,7 +4,7 @@ <Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. <Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. <Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. -<Audio 1> provides a synthetic voice reference for <Subject 1> (S1). +<Audio 1> provides a synthetic voice reference for <Subject 1>. summary: [reference generation] Two referenced adults pass one camera in the referenced cafe while one puppy watches.
Downscaled RGB MAE: 20.3644/255. Audio RMS: A -28.385 dBFS; B -28.247 dBFS.
What changed: Add irrelevant source transcript to retention.
What we checked: Leakage of source sentence.
What we observed: Source-transcript leakage is not assessed without listening or transcription. Do not infer speech correctness from waveform correlation.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -15,7 +15,7 @@ <Subject 3> (appears in [Shot 1]): fully_preserved - Keep the cafe architecture and fixtures while permitting the described camera movement. <Subject 4> (appears in [Shot 1]): fully_preserved - Keep one black camera with the same body and short lens during the handoff. <Subject 5> (appears in [Shot 1]): fully_preserved - Keep one golden puppy with the same coat and proportions during its head movement. -<Audio 1>: reference - use the voice texture for new words, without copying the source sentence. +<Audio 1>: reference - use the voice texture for new words, without copying the source sentence. The source says <d>[English] Please place the camera on the table.</d> detailed_description: A naturalistic live-action scene with neutral skin tones, soft window light and coherent physical motion.
Downscaled RGB MAE: 16.8468/255. Audio RMS: A -28.385 dBFS; B -23.768 dBFS.
What changed: Select Audio 1 versus Audio 2 with identical wiring.
What we checked: Tone track versus voice, proving actual input-label mapping.
What we observed: Both outputs retain the cue tones, not a grossly different voice waveform. However B only changes the final soundscape selection to Audio 2 while definitions and retention still demand Audio 1 and forbid reusing Audio 2. This contradictory control cannot prove whether audio indexing works or is ignored.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -26,7 +26,7 @@ No one speaks. overall_soundscape: -Only <Audio 1> is heard as the complete output soundtrack. +Only <Audio 2> is heard as the complete output soundtrack. non_diegetic_music: N/A
Downscaled RGB MAE: 11.587/255. Audio RMS: A -31.938 dBFS; B -31.97 dBFS.
What changed: Remove alignment preamble; actual keyframe conditioning stays identical.
What we checked: First/last correspondence and intermediate path.
What we observed: With the first-frame image connected in both arms, the opening remains visually close to the source even when alignment prose is removed. Later action differs, including A raising the camera toward the face. This is not a test of whether image conditioning is needed.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,5 +1,3 @@ -Begin the output at time zero with the composition, subjects and object positions of <Picture 1>; then continue the described action. - integrated_multimodal_description: [Shot 1] A naturalistic cafe scene begins with a seated woman holding a black camera on the left and a seated bearded man waiting on the right. One golden puppy sits in front of the man. She passes the camera to him between two and four seconds, he turns it toward himself, and both smile. The view moves slowly closer while preserving the room geometry. At the end he lowers the camera beside the table, matching the supplied ending view when one is connected.
Downscaled RGB MAE: 6.9374/255. Audio RMS: A -53.62 dBFS; B -43.981 dBFS.
What changed: Remove alignment preamble; actual keyframe conditioning stays identical.
What we checked: First/last correspondence and intermediate path.
What we observed: Both connected first/last keyframes remain visually close to their source frames at output frame 0 and frame 242, despite removing the alignment preamble in B. Intermediate gestures differ. Endpoint similarity is visual, not a claim of lossless pixel identity.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,5 +1,3 @@ -Begin at <Picture 1> and arrive at <Picture 2> at the end of this 10.125-second video, following a continuous physical path. - integrated_multimodal_description: [Shot 1] A naturalistic cafe scene begins with a seated woman holding a black camera on the left and a seated bearded man waiting on the right. One golden puppy sits in front of the man. She passes the camera to him between two and four seconds, he turns it toward himself, and both smile. The view moves slowly closer while preserving the room geometry. At the end he lowers the camera beside the table, matching the supplied ending view when one is connected.
Downscaled RGB MAE: 4.1503/255. Audio RMS: A -44.181 dBFS; B -43.267 dBFS.
What changed: Remove alignment preamble; actual keyframe conditioning stays identical.
What we checked: First/last correspondence and intermediate path.
What we observed: Both last-frame-conditioned outputs end close to the supplied final image, while their intermediate actions differ. The ending survives removal of alignment prose because the actual last-frame condition is unchanged.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,5 +1,3 @@ -Infer a plausible opening and converge on <Picture 1> at the end of this 10.125-second video; that image anchors only the ending. - integrated_multimodal_description: [Shot 1] A naturalistic cafe scene begins with a seated woman holding a black camera on the left and a seated bearded man waiting on the right. One golden puppy sits in front of the man. She passes the camera to him between two and four seconds, he turns it toward himself, and both smile. The view moves slowly closer while preserving the room geometry. At the end he lowers the camera beside the table, matching the supplied ending view when one is connected.
Downscaled RGB MAE: 11.7681/255. Audio RMS: A -47.702 dBFS; B -46.008 dBFS.
What changed: Headings only, no references.
What we checked: Action adherence and unsolicited audio.
What we observed: Both T2VA field-name variants produce a cafe camera transfer, with different faces and wardrobe. No reference identity score applies because this mode has no reference-image conditioning.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,8 +1,8 @@ -integrated_multimodal_description: +scene: [Shot 1] A naturalistic cafe scene begins with a seated woman holding a black camera on the left and a seated bearded man waiting on the right. One golden puppy sits in front of the man. She passes the camera to him between two and four seconds, he turns it toward himself, and both smile. The view moves slowly closer while preserving the room geometry. At the end he lowers the camera beside the table, matching the supplied ending view when one is connected. -overall_soundscape: +sounds: Quiet cafe ambience; no speech. -non_diegetic_music: +music: N/A
Downscaled RGB MAE: 30.5495/255. Audio RMS: A -54.966 dBFS; B -47.222 dBFS.
What changed: Remove the physical transition path; keep endpoint targets.
What we checked: Plausible transfer, hand contact and object permanence.
What we observed: Detailed path and concise endpoint prose both transfer the camera between the connected keyframes, but A includes an extra camera-to-eye gesture. The gesture is not explicitly requested, so extra motion is not automatically better adherence.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,7 +1,7 @@ Start with <Picture 1> and reach <Picture 2> at the end of the 10.125-second video. integrated_multimodal_description: -[Shot 1] A naturalistic cafe scene begins with a seated woman holding a black camera on the left and a seated bearded man waiting on the right. One golden puppy sits in front of the man. She passes the camera to him between two and four seconds, he turns it toward himself, and both smile. The view moves slowly closer while preserving the room geometry. At the end he lowers the camera beside the table, matching the supplied ending view when one is connected. +[Shot 1] A seated woman holds a camera on the left at the beginning. A seated man holds it on the right at the end. Preserve the cafe and puppy. overall_soundscape: Quiet room tone; no speech.
Downscaled RGB MAE: 13.5593/255. Audio RMS: A -46.55 dBFS; B -51.753 dBFS.
What changed: Wrong endpoint prose; last-frame conditioning stays fixed.
What we checked: Prompt versus actual conditioning conflict.
What we observed: Both outputs end close to the last-frame condition even when B's prose incorrectly calls it a first-frame alignment. This shows the supplied last-frame condition surviving conflicting wording in this case, not that mode instructions can always be discarded.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,4 +1,4 @@ -The supplied <Picture 1> is only the final frame at 10.125 seconds, not the opening. +The supplied <Picture 1> is the opening frame, not the ending. integrated_multimodal_description: [Shot 1] A naturalistic cafe scene begins with a seated woman holding a black camera on the left and a seated bearded man waiting on the right. One golden puppy sits in front of the man. She passes the camera to him between two and four seconds, he turns it toward himself, and both smile. The view moves slowly closer while preserving the room geometry. At the end he lowers the camera beside the table, matching the supplied ending view when one is connected.
Downscaled RGB MAE: 10.7619/255. Audio RMS: A -44.981 dBFS; B -45.281 dBFS.
What changed: Shared group-photo provenance versus individual references; all images connected.
What we checked: Correct separation of people and puppy.
What we observed: The group-image and individual-subject descriptions both retain the seated pair, puppy and camera transfer. The single group image remains connected in both arms; this does not test removal of visual conditioning.

A at 2 fps · B at 2 fps · Prompt A · Prompt B
--- A +++ B @@ -1,9 +1,9 @@ subject_definitions: -<Subject 1> is the adult woman in <Picture 6>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. -<Subject 2> is the adult man in <Picture 6>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. +<Subject 1> is the adult woman in <Picture 1>; preserve her face, swept-back dark hair and black top, not her portrait background or blue lighting. +<Subject 2> is the adult man in <Picture 2>; preserve his face, short beard, swept-up brown hair and dark cardigan, not the portrait background. <Subject 3> is the cafe in <Picture 3>; use its cream walls, green hanging lamps, dark menu boards, exposed conduits and CAFE sign as an environment, not a fixed viewpoint. <Subject 4> is the single black camera in <Picture 4>; use its compact body, textured grip and short black lens, not the surrounding objects. -<Subject 5> is the single golden retriever puppy in <Picture 6>; use its golden fur, floppy ears and small proportions, not its outdoor background. +<Subject 5> is the single golden retriever puppy in <Picture 5>; use its golden fur, floppy ears and small proportions, not its outdoor background. <Picture 6> shows the woman, man and puppy together in the cafe. summary:
Downscaled RGB MAE: 3.0099/255. Audio RMS: A -25.818 dBFS; B -27.891 dBFS.