# Bet on the banana — raw footage delivery

moneycanbuy.fun may use, change and publish these files without attribution.

## Format and provenance

- Ten separate continuous image-to-video takes generated with **bytedance/seedance-1-pro** from photoreal image-model start stills. The only repository visual references were `toad-head/front.png` and `toad-head/three-quarter.png`, used to design the human-worn plush mascot head.
- Video: H.264, yuv420p, native model vertical **1088 × 1920**, 24 fps, exactly 120 frames and 5.000 s each. The model returned 121 frames; ffmpeg 7.0.2 removed the extra final frame and encoded H.264 with fast-start metadata. No video was upscaled. Each clip has **no audio stream**, as requested for raw footage; this takes precedence over the separate AAC acceptance line.
- Start stills: 941 × 1672 PNG. For model transport each generated still was JPEG encoded, then decoded to the delivered PNG. Pixel comparison found each delivered PNG identical to the JPEG submitted to Seedance. Seedance resampled its input to native video size, so the decoded first MP4 frame is not bitwise identical to the 941 × 1672 source still.
- Music: original MusicGen-generated instrumental, 30.000 s, PCM 16-bit stereo 48 kHz, measured **−14.5 LUFS** integrated and **−1.2 dBTP**. Processing after generation was a 0.09 s front trim to align the beat, a fade with 0.09 s of silent tail to retain exactly 30.000 s, resampling, and gain automation of existing audio only. No musical sound was synthesized or layered in code.
- Contact sheet: 600 × 4210 PNG, 3.75 MiB, showing frames 0, 60, 119 for each final clip and the four cast stills.
- An attempted Kling v2.1 run returned `model discontinued`; none of its outputs were used. The delivered clips all use Seedance 1 Pro. Two later music candidates were also rejected: a 31 s MusicGen take had 878 clipped samples at full scale, and a Lyria 3 take was only 29.13 s with an extended gap. The selected first MusicGen take had neither problem.

## File log

| File | Tool and model/version | Exact prompt | Regenerated and why |
| --- | --- | --- | --- |
| clips/B01.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second single continuous handheld smartphone shot in a real office. The same young woman in cream blouse and olive cardigan sits at her desk, looks at the bright green phone she holds toward camera, subtly shifts her gaze and hand, with natural breathing and small true live-action motions throughout. Keep her face, hair, clothes and body unchanged. Phone screen remains solid flat #00FF00 with all four edges and corners unobstructed, no glare, no UI. Natural daylight, true color, no text, logos, watermarks, or morphing. | None. |
| start/B01.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: The young woman from reference sits at her office desk, holding a smartphone toward camera in one hand and looking at it, calm curious expression. The phone display is prominent and fully visible. Any phone facing the camera shows one solid evenly lit pure chroma green #00FF00 rectangle, with all four edges and corners visible, completely unobstructed, no glare, reflection, icons, UI or text. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| clips/B02.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second continuous candid handheld smartphone footage. The same five office colleagues stand close together laughing and playfully tap and show their green-screen phones toward camera. Each person moves naturally, subtle body sway and spontaneous laughter throughout. Keep each face, hair, skin tone, clothes and body consistent. Every visible phone screen stays flat solid #00FF00 with all four corners visible, no glare, no UI. No text or logos. | None. |
| start/B02.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: The same five colleagues from reference gather close around one desk, laughing and playfully tapping their own smartphones, screens held toward camera. Keep faces individually recognizable and all outfits exactly matching reference. Any phone facing the camera shows one solid evenly lit pure chroma green #00FF00 rectangle, with all four edges and corners visible, completely unobstructed, no glare, reflection, icons, UI or text. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| clips/B03.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second continuous candid handheld smartphone footage. The same young woman alone at desk holds her green-screen phone facing camera, calmly taps the middle of its screen exactly once, then lowers her finger and looks quietly confident. Natural small live-action motion all the way. Keep face, hair, clothes and phone rigid. Green display stays uniformly #00FF00 with all corners visible. No UI, text or logos. | None. |
| start/B03.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: The young woman from reference sits alone at her desk, calm confident expression, one finger about to tap the middle of a smartphone screen held toward the camera. One phone only. Any phone facing the camera shows one solid evenly lit pure chroma green #00FF00 rectangle, with all four edges and corners visible, completely unobstructed, no glare, reflection, icons, UI or text. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| clips/B04.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second continuous candid handheld smartphone footage. The same five colleagues point toward the woman at her desk and laugh, while she stays composed. Natural real-time gestures, smiles and small shifts all five seconds; same recognizable faces and outfits, no morphing. Office daylight, true color. No phones, text or logos. | None. |
| start/B04.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: The five same colleagues from reference stand around the young woman from reference at her desk, pointing at her and laughing. The woman remains composed. All faces and clothes match references. No phones visible. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| clips/B05.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second continuous candid handheld smartphone footage in morning office. Elevator doors finish opening and the same banana-suited man walks out in real time carrying a plain blank rigid coffee cup. His face stays visible and his costume unchanged. The same woman and five colleagues freeze, turn heads and stare in surprise. Natural motion all the way, no slow motion. No text, logos or brands. | None. |
| start/B05.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: Morning office. Open elevator doors at center; the exact banana-costumed man from reference takes a step out holding a completely blank white rigid paper coffee cup. He wears the same full yellow costume with his face visible. The young woman and five colleagues, same faces and clothes as their references, freeze and stare at him in surprise. No phone screens visible. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| clips/B06.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second continuous candid handheld smartphone footage. The same man in full yellow banana costume works at a desk, face visible, looking down and typing on keyboard, while the same five colleagues film him with phones and react, laughing with natural real-time movement. One completely blank yellow sticky note stays rigid on the monitor bezel. The monitor is switched off black. All phone screens face camera and remain flat solid #00FF00, all four corners visible, no glare, UI, text or logos. Keep every face, costume and clothes unchanged. | None. |
| start/B06.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: The exact banana-costumed man from reference sits working at a desk, face visible, hands on a plain keyboard. The five same colleagues film him with smartphones held toward camera. A completely blank yellow sticky note is on the bezel of his switched-off dark monitor. Same costume, faces and clothes as reference. Screens on the phones only are chroma green. Any phone facing the camera shows one solid evenly lit pure chroma green #00FF00 rectangle, with all four edges and corners visible, completely unobstructed, no glare, reflection, icons, UI or text. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| clips/B08.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second continuous candid handheld smartphone footage. The same young woman in cream blouse and olive cardigan looks at the fully visible solid green #00FF00 phone screen, then clenches and raises her fist in triumph with a real spontaneous grin. Natural body and hand motion throughout, no slow motion. Keep her face, curly hair, clothes and phone unchanged. Phone display flat bright green, no glare, all four corners visible, no UI, text or logos. | None. |
| start/B08.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: The young woman from reference at her desk looks at one smartphone held fully toward camera, smiling with a clenched celebratory fist beside her face, poised for triumph. Same face, hair, cream blouse and olive cardigan as reference. Any phone facing the camera shows one solid evenly lit pure chroma green #00FF00 rectangle, with all four edges and corners visible, completely unobstructed, no glare, reflection, icons, UI or text. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| clips/B09.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second continuous real-time handheld smartphone footage between the supplied exact start and end frames. Human performer in the same giant green fabric toad plush head stays fully in frame and front-facing throughout. Its heavy HALF-CLOSED cream eyes, flat smirk, cheek spots, dark hoodie, gold chain and '$' chest tag never change and remain visible. The clear rigid box tips and a large pile of blank gold coins tumbles onto the same woman's desk. Colleagues stare with jaws dropped, same faces and clothes. Natural daylight. No text, logos, morphing, change of costume, or camera pan. One take, real motion all the way. | Regenerated after first take moved the toad partly out of frame; final run used a generated end still. |
| start/B09.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: The exact giant plush toad-headed HUMAN from reference, in plain dark hoodie with visible gold chain and one gold '$' tag on chest, stands by the young woman's desk. He tips a transparent rigid locked box of completely blank gold coins, coins beginning to spill in a big pile onto her desk. The same woman and five colleagues stare, shocked, jaws dropped. Same faces and clothes as references. Toad head faces camera at three-quarter angle, heavy half-closed cream eyes, flat smirk, cheek spots. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| clips/B10.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second continuous real-time handheld smartphone footage between supplied exact start and end frames. Human performer in the SAME giant green sewn toad plush head stays fully visible front three-quarter, with heavy HALF-CLOSED cream eyes, flat smirk and cheek spots absolutely unchanged; never wide-open eyes. Dark hoodie, gold chain and single '$' chest tag remain clearly visible. The transparent rigid box tips; plain blank gold coins fall into the metal bowl of REAL LIVE FIRE, orange flames flare upward. Preserve box, bowl, head, hands and background consistency. Natural true-color office light, no text, logos, morphing or other symbols. One take, genuine motion all the way. | Regenerated after first take widened the toad eyes; final run used a generated end still. |
| start/B10.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: The exact giant plush toad-headed HUMAN from reference, in plain dark hoodie with visible gold chain and one gold '$' tag on chest, front three-quarter view, pours completely blank gold coins from a transparent rigid locked box into a metal bowl of REAL LIVE FIRE on a fire-safe office tabletop. Yellow-orange flames already visible, realistic heat and flicker, no graphics. Keep the toad's heavy half-closed cream eyes, flat smirk and cheek spots exactly like reference. No other people. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| clips/B11.mp4 | Replicate / bytedance/seedance-1-pro (a5fd550893da3b6f67997812759065652454ddaca10e96b83b59cbae1814cb36) | Five-second continuous candid handheld smartphone footage. The same yellow banana-costumed man and same young woman in cream blouse and olive cardigan complete one joyful high-five, their palms touch once, then their hands separate as they laugh and smile. Behind them the same five colleagues keep looking down toward the floor, disappointed. Everyone moves naturally in real time. Keep all faces, hair, clothes, banana costume and body shapes consistent. No phones, text, brands, logos or watermarks. | None. |
| start/B11.png | OpenAI image_gen built-in (underlying model/version not exposed) | Photoreal live-action candid vertical 9:16 smartphone footage START STILL, same recognizable fictional people and exact clothes as cast references. Ordinary modern unbranded open-plan office, natural daylight, true colors, real skin and cloth texture, slightly imperfect phone framing, no cinematic color grade. No letters, numbers, logos, brand marks, printed packaging or readable text anywhere. Blank walls and posters. Computer monitors switched off dark glass with no text reflection. All props rigid and physically plausible. No watermarks. Scene: The same banana-costumed man and young woman from references face one another and raise their hands about to high-five, happy smiles. Behind them, the same five colleagues from reference look down at the floor, disappointed. All original faces, hair and clothes match references. No phones visible. | No regeneration of delivered still; PNG is the exact pixel decode of the JPEG image submitted to the video model. |
| track/track.wav | Replicate / meta/musicgen stereo-large (671ac645ce5e552cc63a54a2bbff63fcf798043055d2dac5fc9e36a837eedcfb) | Original instrumental only, no voices. Light plucky office comedy music at 120 BPM, 4/4, first downbeat immediately. Marimba and pizzicato strings, dry drums and bass. Bigger groove at 9.5 seconds, quiet suspense from 15 to 17, triumphant groove from 17, final hit at 28.5 seconds. No existing melody. | First model generation selected; a second 31 s candidate was rejected because its music faded too early. The selected waveform was trimmed, end faded, resampled and gain automated. |
| track/beats.json | NumPy 2.5.4 / SciPy 1.18.1 onset analysis | N/A — beat grid and accent times measured from track/track.wav. | None. |
| contact-sheet.png | Pillow 13.0.0 / ffmpeg 7.0.2 | N/A — arranged frames 0, 60 and 119 of each final clip, plus the four cast reference stills. | Rebuilt after B09 and B10 were regenerated. |
| README.md | Hand-authored delivery log | N/A — documentation. | Updated after final validation. |

## Cast reference prompts

These generated cast stills appear only in `contact-sheet.png`. They were used as image references for the scene start stills; the toad cast still also used the two specified repository reference images.

- **bettor:** Photorealistic candid smartphone CAST REFERENCE portrait, vertical 9:16. Fictional young adult woman office worker, about 27, medium brown skin, shoulder-length dark curly hair, warm brown eyes, distinctive small beauty mark beside her left cheek, natural face, slim average build. Plain cream blouse, dark olive cardigan and dark trousers, no jewelry. Neutral expression, clear recognizable face and outfit, body visible head to knee. Ordinary open-plan office, natural daylight, true colors, real skin and fabric texture, casual imperfect phone camera, no cinematic grading. Absolutely no text, letters, symbols, brands or logos anywhere.
- **banana:** Photorealistic candid smartphone portrait, vertical 9:16. A fictional adult man, around 32, olive skin, short dark brown hair, brown eyes, natural face and average build, wearing a full padded bright yellow banana costume with his face visible through the opening, no letters, symbols, brands or logos. Standing in an ordinary unbranded office, natural daylight, true skin and fabric texture, casual imperfect phone camera framing, no cinematic grading. CAST REFERENCE image with clear face and full costume. No text anywhere.
- **colleagues:** Photorealistic candid smartphone CAST REFERENCE group portrait, vertical 9:16. Five distinct ordinary fictional office adults, all clearly visible head to knee: woman with fair skin and shoulder-length straight auburn hair in plain blue shirt; Black man with close-cropped hair in plain gray sweater; East Asian woman with chin-length black bob in plain lavender blouse; South Asian man with wavy black hair and neat beard in plain brown shirt; white woman with short blond pixie cut in plain green sweater. Same outfits will recur. Diverse natural faces, no famous people. Standing side by side in an ordinary open-plan office, natural daylight, true colors and real skin/fabric texture, imperfect casual phone framing, no cinematic grading. No letters, text, symbols, brands or logos anywhere.
- **toad:** Use case: identity-preserve photoreal prop conversion. Input images are two exact character identity references: first is front, second is three-quarter. Generate a realistic human performer wearing a big physical FABRIC PLUSH MASCOT HEAD based exactly on the referenced toad: flat green broad head, heavy half-closed upper lids over cream eyes, tiny nostrils, wide flat smirk, dark cheek spots. Never a generic frog or round open eyes. Human body in a plain dark charcoal hoodie. A chunky gold chain with a gold chest tag bearing exactly one dollar symbol '$', visible. Full body, front three-quarter view, in an unbranded office, natural light, sewn plush texture, candid smartphone photo, vertical 9:16. No other letters, numbers, text, logos or symbols anywhere.

## Regenerated end still prompts

The B09 and B10 final video generations used their respective start stills plus generated end-frame references. These end images are intermediate inputs and are not delivered separately.

- **B09:** Edit the first reference scene as the END FRAME of the same unbroken real office shot, vertical 9:16 photoreal smartphone photo. Preserve every person, face, hair, clothing, the exact plush toad mascot head, its heavy HALF-CLOSED cream eyes, flat smirk and cheek spots. Keep toad fully in frame facing camera, with its gold chain and gold '$' tag clearly visible on chest. Its clear rigid coin box is now almost empty, tilted just over the desk. The blank gold coins have formed a large pile on the young woman's desk; a few are still falling. The same woman and colleagues remain startled, jaws dropped. Natural daylight, true colors, no text, letters, numbers, logos, other symbols, reflections or morphing. Do not change camera position or composition significantly.
- **B10:** Edit the first reference scene as the END FRAME of the same unbroken real office shot, vertical 9:16 photoreal smartphone photo. Preserve exact human in giant physical fabric plush TOAD head, heavy HALF-CLOSED cream eyes at all times, flat smirk, dark cheek spots, same dark hoodie and clearly visible gold chain with '$' tag. Toad stays fully in frame facing camera, never generic frog. Transparent rigid coin box now nearly empty, tipped over metal bowl. Pile of plain blank gold coins in bowl, real orange live fire flames flare moderately higher around pile. Same background, camera position and composition, daylight, true colors. No other letters, numbers, logos, symbols or text.

## Music timing

The `beats.json` grid is 120 BPM in 4/4, from 0.000 through 29.500 s at 0.500 s intervals; downbeats are every 2.000 s. The accent times below were checked against the final WAV's spectral onsets or, for the drop at 15.000 s, the applied gain edge.

| Requested second | Measured second | Offset (s) | Cue |
| ---: | ---: | ---: | --- |
| 2.5 | 2.496 | -0.004 | hit |
| 5.0 | 5.006 | +0.006 | hit |
| 7.5 | 7.516 | +0.016 | hit |
| 9.5 | 9.516 | +0.016 | drop |
| 12.5 | 12.501 | +0.001 | hit |
| 15.0 | 15.000 | +0.000 | drop |
| 17.0 | 17.006 | +0.006 | hit |
| 19.0 | 19.001 | +0.001 | coin |
| 22.5 | 22.501 | +0.001 | fire |
| 25.5 | 25.516 | +0.016 | hit (clap cue) |
| 28.5 | 28.501 | +0.001 | final |

The 15–17 s section is lowered substantially by permitted gain automation. The final cue is a gain lift of a generated musical transient. The exact timbre of named cues, such as fire and coin ticks, remains model generated and may be less literal than a designed sound effect.

## Rejection-list review

1. Letters, numbers or logos: no readable marks were found in the reviewed start/middle/end frames of all ten final clips and the start stills; the toad's `$` tag is visible. Review of sampled frames cannot certify every decoded frame.
2. Green surfaces: phones appear uniformly bright green and unobstructed in the reviewed frames, but **pixel-exact #00FF00 is not achieved** by the video model and H.264. A sampled B01 green region decoded with mean RGB about (18, 253, 17) and local variation. These are raw model takes, so no screen replacement was composited.
3. Faces, hands and props: cast identities and clothing remain recognizable in reviewed frames; the rigid cup, box, sticky note and coins retained their shapes in sampled views. Fine-frame morphing remains a model risk.
4. Motion: start/middle/end frames show continuing real-time motion in every clip; all are model-generated video, with no slideshow, still pan, code animation, loop or slow motion.
5. Toad: B09 and B10 were regenerated after initial visual failures; the final sampled frames retain the specific flat head, half-closed lids, cheek spots, chain and `$` tag, with no rear view.
6. Format: all ten final MP4 files decode as H.264/yuv420p at 1088 × 1920, 24 fps, 120 frames, exactly 5.000 s, with no audio. No watermark was observed in sampled frames.
7. Brand and likeness: cast images were generated as fictional ordinary adults and no recognizable brand or famous likeness was observed in sampled frames.
8. Music: model-generated instrumental WAV is 30.000 s and was not composed from code or loops. VAD-assisted Whisper returned no words, and Gemini 3 Flash's audio check judged that it contains no voices, words, singing or vocal chops. A basic Whisper run produced one whole-track “Thank you” hypothesis, so automatic vocal checks are not unanimous.

## Verification limits

The screen color limitation above is material for keying. Visual checks used the first, middle and final frames of each completed clip plus full-length playback metadata; they do not prove every frame is free of tiny text, subtle morphing or green spill. The music cues were measured from the waveform; audio-model checks support the no-vocal requirement but cannot certify it absolutely.