The Reference Sheet — A Building Made of the Car
Find an accomplice
SEEDANCE 2.5 · VIDEO PROMPT · 9:16 · @IMAGE 1 + @IMAGE 2
FORMAT: 30s / 9:16 vertical / luxury fashion film into automotive ASMR drive / fast-cut opening, then long takes / no music STYLE: Hollywood live-action realism, ARRI Alexa LF, anamorphic, fine 35mm grain, warm low late-afternoon sun, long soft shadows, natural light only, shallow depth of field on the man, editorial fashion grading in cream, sand, caramel and white. One visual identity, no style mixing. REFERENCES: @Image 1 is a six-panel sheet of ONE white wide-body sports coupe and ONE location. It defines only: the car (white body, wide arches, large white swan-neck rear wing, white multi-spoke wheels, full-width red rear light bar, cream and caramel leather interior, caramel steering wheel) and the environment (white curved architecture, facade lettering exactly as shown, horizontal red light strip, two white arched hangar entrances, sand-colored concrete plaza, warm sky). Ignore the grid layout. The panels are NOT keyframes, shot order or angles to copy. Exactly one car, no collage. @Image 2 defines the hero's face and identity ONLY: dark brown tight curly medium-length hair, black square-framed sunglasses, short mustache and light beard, warm skin tone, relaxed half-smile. Do not inherit his grey t-shirt, food or restaurant background. WARDROBE (fixed): cream heavyweight boxy tee; open chocolate-brown suede overshirt, cuffs pushed up; cream wide-leg pleated wool trousers; cream-and-brown suede New Balance 9060 sneakers with a chunky layered sole; matte black square sunglasses never removed. VIDEO PROMPT: [Stage 1 — approx 0–5s] FASHION CUT. Six hard cuts of about 0.8s, no transitions, each a different lens, all on the plaza, white car in the far background. Cut 1: ground-level macro, a sneaker lands on concrete, trouser hem swings. Cut 2: 100mm close-up of the suede sleeve, hand pushing the cuff up. Cut 3: tight profile of the sunglasses, the white car reflected in the lens. Cut 4: 35mm handheld from behind, curls catching sun, car soft ahead. Cut 5: waist-level tracking of the trousers. Cut 6: high oblique drone at about 45 degrees, the man a small figure crossing toward the car, long shadow leading him. End state: man mid-plaza, walking left to right toward the driver door, car about 15 meters away. [Stage 2 — approx 5–11s] Continue from mid-plaza. Camera: one continuous drone shot, diagonal top-down at about 45 degrees, descending and drifting to hold man and car in one frame. He walks at a steady unhurried pace, hands relaxed. At 9s the drone is about 3 meters off the ground behind him as he reaches the driver door. End state: he stands at the driver door, right hand on the handle, sunglasses on. [Stage 3 — approx 11–18s] ASMR INTERIOR. Continue from hand on handle. Camera: restrained handheld, breath-level micro movement only, no gimbal smoothness. Shot A: close-up of the handle, the door unlatches and swings open. Shot B: from the passenger seat, he lowers into the cream seat, door pulls shut with one solid close. Shot C: macro of his hands settling on the caramel wheel, thumbs on the stitching. Shot D: macro of the seatbelt buckle clicking home. Shot E: macro of fingers on the ignition, one press at 17s; the dashboard wakes. End state: engine at idle, both hands on the wheel, sunglasses on, door closed. [Stage 4 — approx 18–26s] DRIVE. Continue from idle. Shot A: low static ground-level shot beside the rear wheel, the car pulls out of the plaza left to right, rear wing and red light bar crossing frame. Shot B: drone chase from a high oblique 45-degree angle behind, the car leaving the plaza onto a long empty two-lane asphalt road through pale sand terrain, the drone descending as the car accelerates; speed shown by road markings streaming past. Shot C: interior over-the-shoulder from the passenger headrest, his hand shifting, road unrolling through the windshield, sunlight sliding across the sunglasses. End state: car at speed on the empty road, no other vehicles. [Stage 5 — approx 26–30s] CLOSING. One continuous drone shot: from a rear-following position the drone rises and slows while the car continues away, becoming small on the straight road with its long shadow, the architecture faint on the horizon. Hold the final frame for the last second, car still moving. Final state: one white car receding on an empty road, no text. AUDIO: No music at any point. Diegetic sound only, close-miked, ASMR clarity. Stage 1: <crisp sneaker steps on concrete, suede rustle, light dry wind, each cut lands on a footstep>. Stage 2: <footsteps growing closer, soft wind>. Stage 3: <door latch, leather creak as he sits, one deep solid door close, hands on stitched leather, seatbelt click, at 17s a single ignition press then a raw naturally aspirated flat-six cold start bark settling into a textured idle with low-frequency pressure>. Stage 4: <tires on asphalt, engine climbing the rev range, one clean upshift, exhaust echoing off the architecture then opening into wide space, wind rising>. Stage 5: <engine fading into distance, wind takes over, final second near silent>. No voiceover. CONSISTENCY: Exactly one man, the hero from @Image 2. Exactly one car, the white coupe from @Image 1. Keep face, curly hair, black square sunglasses, wardrobe, car color, white wheels, rear wing, red light bar and interior unchanged in every shot. Sunglasses stay on for all 30 seconds, including inside the car. Man walks left to right; car exits left to right, then away from camera. Sun low from the left throughout. No other people, vehicles or graphics. NEGATIVE PROMPT: no music, no voiceover, no subtitles, no on-screen text, no watermarks, no split-screen, no grid, no panel borders, no collage, no second car, no duplicate man, no face distortion, no identity change, no sunglasses removed, no wardrobe change, no grey t-shirt, no hat, no warped wheels, no morphing car body, no extra rear wings, no cartoon, no cheap CGI, no plastic skin, no flat lighting, no smooth gimbal in Stage 3, no night, no crowd.
Other prompts in this tutorial
NANO BANANA PRO · JSON PROMPT · 16:9
{ "task": "Photorealistic 6-panel automotive editorial grid. One white Porsche 911 GT3 RS photographed from multiple angles in front of a colossal white building whose architecture is derived from the same car's design language. Two panels show the interior. All six panels are real photographs from one shoot, same car, same location, same light.", "references": { "image_1": "Car identity and form reference. Read the exact body shape, wing, wheels, stance, ride height, white paint finish and all proportions from this image. Do not reinterpret the car. Do not add or remove aero parts. The warm beige-peach tonality of this image is the color world of the whole grid." }, "layout": { "format": "single 16:9 image divided into a clean 2x3 grid, three panels per row", "gutters": "thin uniform off-white gutters between panels, equal width, no rounded corners, no drop shadows, no labels, no numbers", "consistency": "identical car, identical building, identical time of day and light direction in every panel" }, "environment": { "building": "A colossal white monolithic building designed as a piece of Porsche GT3 RS architecture. The facade is smooth white gloss-ceramic panels that mirror the car's paint. The wide rear wheel arches of the car become two enormous curved white volumes at the building's base. The GT3 RS swan-neck rear wing becomes a cantilevered white roof canopy, nine storeys up, supported by two slim angled struts. The car's side air intakes become tall louvered black window slots cut into the facade. A single continuous horizontal red LED light bar runs across the building like the car's rear taillight strip. The word 'PORSCHE' appears once, in the brand's real letterspaced capitals, as a very large brushed-metal sign on the upper facade, and the crest appears once, small, at the entrance. The building is scaled so the car looks small in front of it.", "ground": "A vast empty plaza of smooth pale sand-colored concrete, wide seamless slabs, faint tire dust marks, a few subtle expansion joints. No other cars, no people.", "sky": "Warm hazy late-afternoon sky in soft beige and peach, no clouds, matching the tonality of [Image 1]." }, "vehicle": { "identity": "[Image 1] car, unchanged. White paint with real gloss depth and slight orange-peel reflection, not a flat render white.", "interior": "Premium fully leather-trimmed cabin in two tones: cognac brown and warm cream. Cream leather sport bucket seats with cognac bolsters, fine cognac contrast stitching, perforated center panels, a small embossed Porsche crest on the headrests. Cognac leather dashboard top with cream lower dashboard, cream leather door panels with cognac armrests, cognac Alcantara-free full leather steering wheel with a yellow 12 o'clock marker, carbon fiber trim on the center console, matte black metal paddle shifters, digital instrument cluster switched on with the tachometer glowing. Cream Alcantara headliner. Every surface shows real leather grain, slight natural creasing on the seat cushions, and honest reflections." }, "panels": [ { "panel": 1, "position": "top-left", "shot": "Wide establishing shot, front three-quarter view of the car, camera at knee height 24mm, the car parked dead center of the plaza with the entire building rising behind it. Car occupies the lower third of the frame, building fills the rest." }, { "panel": 2, "position": "top-center", "shot": "Rear three-quarter view matching the angle of [Image 1], camera at hip height 35mm, the rear wing of the car visually aligned under the building's wing-shaped roof canopy so the two shapes echo each other." }, { "panel": 3, "position": "top-right", "shot": "Pure side profile, camera exactly perpendicular to the car at wheel-hub height, 85mm from far back for compressed perspective, the horizontal red light bar of the building running straight across the background behind the car." }, { "panel": 4, "position": "bottom-left", "shot": "Direct low front view, camera 30cm above the ground, 24mm, the car's nose filling the lower frame, the building's two arch-shaped base volumes framing the car on both sides." }, { "panel": 5, "position": "bottom-center", "shot": "Interior, driver's cockpit shot from the open driver door at seat height, 24mm, showing the cream and cognac bucket seat, the steering wheel with yellow marker, the glowing cluster, the carbon center console. Warm light from outside falls across the leather. The white building is softly visible through the windshield." }, { "panel": 6, "position": "bottom-right", "shot": "Interior detail, close 50mm shot from the passenger side looking across both seats, focus on the cognac contrast stitching and perforated cream leather of the seat bolster, the door panel and cognac armrest in the foreground, shallow depth of field, the driver's seat softly blurred behind." } ], "camera_and_light": { "camera": "Phase One IQ4 medium format for exterior panels, Canon 5D Mark IV for interior panels, all handheld on a real shoot, exteriors at f/8 with deep focus, interiors at f/2.8.", "light": "Low warm sun from camera-left at roughly 20 degrees above the horizon, long soft shadows stretching to the right across the plaza, warm highlights and slightly cooler shadows, gentle atmospheric haze softening the top of the building.", "grade": "Muted warm grade in beige, peach and cream, whites kept slightly warm not clinical, blacks lifted a touch, fine medium-format sensor grain visible in the sky and shadows, subtle halation on the red light bar and on chrome edges." }, "quality_constraints": { "realism": "These are real photographs, not renders. Real paint reflections with slight distortion, tire sidewall texture with visible lettering, brake dust on the wheel barrels, faint dust on the plaza, real leather grain and creasing. No CGI smoothness, no perfectly even studio lighting, no floating car, tires sit with real contact shadow.", "banned": "no text other than 'PORSCHE' and the crest, no watermark, no panel numbers, no captions, no extra cars, no people, no reflections revealing a studio, no duplicated or mirrored panels, no fantasy elements, no glowing edges, no digital art look", "identity_lock": "the car in all four exterior panels must be pixel-consistent with [Image 1] in shape, wing, wheels and stance" } }
Tutorial
DENIZ AKKABAK · SEEDANCE 2.5 TUTORIAL · @DENIZ.AKKABAK
MADE IN INVIDEO · invideo.io/i/DENIZ-ig

NANO BANANA PRO → SEEDANCE 2.5
The
Cold Start
One car photo becomes a Porsche world. One portrait walks into it. Twenty-two seconds later, a flat-six barks awake — and there is no music anywhere.
A TUTORIAL BY DENIZ AKKABAK · @DENIZ.AKKABAK
The film on the left is the finished output. Open this document in Google Docs to watch it play.
| PIPELINENano Banana Pro → Seedance 2.5 | FORMAT9:16 vertical · 22s cut | SOUNDDiegetic only · no music | REFERENCESTwo images · one job each |
Most AI car films fail before the first frame. The model is handed one hero shot and asked to invent a whole world around it — and every angle it invents disagrees with the last. This film flips the order. Nano Banana Pro builds the world first, as a six-panel editorial contact sheet: four exteriors, two interiors, one car, one building that is literally the car's design language scaled up nine storeys. Then Seedance 2.5 is handed that sheet and told the one thing most people forget to say — the panels are not keyframes. They are the place. What the model shoots there is the prompt's business: a fashion cut, a drone approach, an ASMR interior, a drive, and a cold start with no music to hide behind.
THE PIPELINE AT A GLANCE
- 01
Build the world in one image
Nano Banana Pro. A single car photo goes in. A JSON prompt asks for a 2×3 grid — same car, same building, same light — with the building derived from the car itself. The sheet that comes out is the location scout for everything that follows.
- 02
Cast yourself, then drive
Seedance 2.5 inside invideo. Two references: the sheet and your own portrait, each with exactly one job. A five-stage, timecoded prompt with a fixed wardrobe, an end state per stage and diegetic sound only produces the finished 9:16 film.
STEP 01 — NANO BANANA PRO
The Reference Sheet — A Building Made of the Car
The trick in Step 01 is architectural, in both senses. The JSON asks the model to read the car — body, wing, wheels, stance, the warm beige-peach tonality of the photo — and then to reinterpret every one of those features as a building: the wide rear arches become two curved white volumes at the base, the swan-neck wing becomes a cantilevered roof canopy, the side intakes become louvered black window slots, the taillight becomes a continuous red LED bar across the facade. Because the location is derived from the vehicle, the two never fight for the frame. And because the request is a grid, you get four exterior angles and two interior details in one generation — a whole shoot, one image.
REFERENCE INPUT — ATTACH THIS IMAGE

NANO BANANA PRO · JSON PROMPT · 16:9
{ "task": "Photorealistic 6-panel automotive editorial grid. One white Porsche 911 GT3 RS photographed from multiple angles in front of a colossal white building whose architecture is derived from the same car's design language. Two panels show the interior. All six panels are real photographs from one shoot, same car, same location, same light.", "references": { "image_1": "Car identity and form reference. Read the exact body shape, wing, wheels, stance, ride height, white paint finish and all proportions from this image. Do not reinterpret the car. Do not add or remove aero parts. The warm beige-peach tonality of this image is the color world of the whole grid." }, "layout": { "format": "single 16:9 image divided into a clean 2x3 grid, three panels per row", "gutters": "thin uniform off-white gutters between panels, equal width, no rounded corners, no drop shadows, no labels, no numbers", "consistency": "identical car, identical building, identical time of day and light direction in every panel" }, "environment": { "building": "A colossal white monolithic building designed as a piece of Porsche GT3 RS architecture. The facade is smooth white gloss-ceramic panels that mirror the car's paint. The wide rear wheel arches of the car become two enormous curved white volumes at the building's base. The GT3 RS swan-neck rear wing becomes a cantilevered white roof canopy, nine storeys up, supported by two slim angled struts. The car's side air intakes become tall louvered black window slots cut into the facade. A single continuous horizontal red LED light bar runs across the building like the car's rear taillight strip. The word 'PORSCHE' appears once, in the brand's real letterspaced capitals, as a very large brushed-metal sign on the upper facade, and the crest appears once, small, at the entrance. The building is scaled so the car looks small in front of it.", "ground": "A vast empty plaza of smooth pale sand-colored concrete, wide seamless slabs, faint tire dust marks, a few subtle expansion joints. No other cars, no people.", "sky": "Warm hazy late-afternoon sky in soft beige and peach, no clouds, matching the tonality of [Image 1]." }, "vehicle": { "identity": "[Image 1] car, unchanged. White paint with real gloss depth and slight orange-peel reflection, not a flat render white.", "interior": "Premium fully leather-trimmed cabin in two tones: cognac brown and warm cream. Cream leather sport bucket seats with cognac bolsters, fine cognac contrast stitching, perforated center panels, a small embossed Porsche crest on the headrests. Cognac leather dashboard top with cream lower dashboard, cream leather door panels with cognac armrests, cognac Alcantara-free full leather steering wheel with a yellow 12 o'clock marker, carbon fiber trim on the center console, matte black metal paddle shifters, digital instrument cluster switched on with the tachometer glowing. Cream Alcantara headliner. Every surface shows real leather grain, slight natural creasing on the seat cushions, and honest reflections." }, "panels": [ { "panel": 1, "position": "top-left", "shot": "Wide establishing shot, front three-quarter view of the car, camera at knee height 24mm, the car parked dead center of the plaza with the entire building rising behind it. Car occupies the lower third of the frame, building fills the rest." }, { "panel": 2, "position": "top-center", "shot": "Rear three-quarter view matching the angle of [Image 1], camera at hip height 35mm, the rear wing of the car visually aligned under the building's wing-shaped roof canopy so the two shapes echo each other." }, { "panel": 3, "position": "top-right", "shot": "Pure side profile, camera exactly perpendicular to the car at wheel-hub height, 85mm from far back for compressed perspective, the horizontal red light bar of the building running straight across the background behind the car." }, { "panel": 4, "position": "bottom-left", "shot": "Direct low front view, camera 30cm above the ground, 24mm, the car's nose filling the lower frame, the building's two arch-shaped base volumes framing the car on both sides." }, { "panel": 5, "position": "bottom-center", "shot": "Interior, driver's cockpit shot from the open driver door at seat height, 24mm, showing the cream and cognac bucket seat, the steering wheel with yellow marker, the glowing cluster, the carbon center console. Warm light from outside falls across the leather. The white building is softly visible through the windshield." }, { "panel": 6, "position": "bottom-right", "shot": "Interior detail, close 50mm shot from the passenger side looking across both seats, focus on the cognac contrast stitching and perforated cream leather of the seat bolster, the door panel and cognac armrest in the foreground, shallow depth of field, the driver's seat softly blurred behind." } ], "camera_and_light": { "camera": "Phase One IQ4 medium format for exterior panels, Canon 5D Mark IV for interior panels, all handheld on a real shoot, exteriors at f/8 with deep focus, interiors at f/2.8.", "light": "Low warm sun from camera-left at roughly 20 degrees above the horizon, long soft shadows stretching to the right across the plaza, warm highlights and slightly cooler shadows, gentle atmospheric haze softening the top of the building.", "grade": "Muted warm grade in beige, peach and cream, whites kept slightly warm not clinical, blacks lifted a touch, fine medium-format sensor grain visible in the sky and shadows, subtle halation on the red light bar and on chrome edges." }, "quality_constraints": { "realism": "These are real photographs, not renders. Real paint reflections with slight distortion, tire sidewall texture with visible lettering, brake dust on the wheel barrels, faint dust on the plaza, real leather grain and creasing. No CGI smoothness, no perfectly even studio lighting, no floating car, tires sit with real contact shadow.", "banned": "no text other than 'PORSCHE' and the crest, no watermark, no panel numbers, no captions, no extra cars, no people, no reflections revealing a studio, no duplicated or mirrored panels, no fantasy elements, no glowing edges, no digital art look", "identity_lock": "the car in all four exterior panels must be pixel-consistent with [Image 1] in shape, wing, wheels and stance" } }
OUTPUT

STEP 02 — SEEDANCE 2.5 INSIDE INVIDEO
The Film — Two References, Five Stages, No Music
Step 02 is a casting session followed by a shoot. The sheet arrives as @Image 1 and is immediately narrowed to two jobs — the car and the place — with the grid itself explicitly waived. Your own portrait arrives as @Image 2 and is narrowed even harder: face and identity only, with the grey t-shirt, the food and the restaurant background refused by name. Then a WARDROBE block dresses the hero from scratch, and the prompt walks Seedance through five stages that each end with a written end state, so the next stage can begin with the words continue from.
Look at the shape of the thirty seconds before you read a single shot: six cuts in five seconds, then one drone shot, then five macro shots with no gimbal, then three drive angles, then a single shot that simply lets the car go. Fast, slow, close, wide, gone. That rhythm is written into the FORMAT line before anything is described — the stage map on the next page is just the FORMAT line unpacked.
REFERENCE INPUTS — ATTACH IN THIS ORDER
REFERENCE
DEFINES
WAIVED IN THE PROMPT

@Image 1 the six-panel sheet
The car — white body, wide arches, swan-neck wing, white wheels, the red rear light bar, the cream-and-caramel cabin.
The place — white curved architecture, the facade lettering, the horizontal red strip, two arched hangar entrances, the sand-colored plaza.
The grid. The panel order. The camera angles. The prompt says it plainly: the panels are NOT keyframes — they are a location scout, not a storyboard.

@Image 2 your portrait
Face and identity ONLY — dark tight curls, black square sunglasses, short mustache and light beard, warm skin, the half-smile.
Everything around the face. The grey t-shirt, the food, the restaurant — all explicitly refused, so the WARDROBE block can dress him from scratch.
THE STAGE MAP — HOW 30 SECONDS ARE SPENT
| TIMECODE | STAGE · CAMERA · END STATE | SOUND | |
| 1 | 0–5s | Fashion cutSix hard cuts of about 0.8s — sneaker, suede cuff, sunglasses with the car in the lens, curls from behind, trousers, then a 45° drone. End state: mid-plaza, walking left to right. | Crisp sneaker steps, suede rustle, dry wind — each cut lands on a footstep. |
| 2 | 5–11s | The approachOne continuous drone, diagonal top-down, descending to 3 m behind him at 9s. End state: right hand on the driver handle, sunglasses on. | Footsteps growing closer, soft wind. |
| 3 | 11–18s | ASMR interiorRestrained handheld, breath-level only. Handle → he lowers into the cream seat → hands on the caramel wheel → seatbelt → ignition at 17s. End state: idle, both hands on the wheel. | Latch, leather creak, one deep door close, seatbelt click, then the flat-six cold start bark. |
| 4 | 18–26s | DriveLow static beside the rear wheel as it pulls out left to right; 45° drone chase onto the empty two-lane road; over-the-shoulder from the passenger headrest. End state: at speed, no other vehicles. | Tires on asphalt, revs climbing, one clean upshift, exhaust off the architecture then into open space. |
| 5 | 26–30s | ClosingOne drone shot rising and slowing while the car keeps going; final frame held, car still moving. Final state: one white car receding, no text. | Engine fading, wind takes over, the last second near silent. |
SEEDANCE 2.5 · VIDEO PROMPT · 9:16 · @IMAGE 1 + @IMAGE 2
30s / 9:16 vertical / luxury fashion film into automotive ASMR drive / fast-cut opening, then long takes / no music
Hollywood live-action realism, ARRI Alexa LF, anamorphic, fine 35mm grain, warm low late-afternoon sun, long soft shadows, natural light only, shallow depth of field on the man, editorial fashion grading in cream, sand, caramel and white. One visual identity, no style mixing.
@Image 1 is a six-panel sheet of ONE white wide-body sports coupe and ONE location. It defines only: the car (white body, wide arches, large white swan-neck rear wing, white multi-spoke wheels, full-width red rear light bar, cream and caramel leather interior, caramel steering wheel) and the environment (white curved architecture, facade lettering exactly as shown, horizontal red light strip, two white arched hangar entrances, sand-colored concrete plaza, warm sky). Ignore the grid layout. The panels are NOT keyframes, shot order or angles to copy. Exactly one car, no collage. @Image 2 defines the hero's face and identity ONLY: dark brown tight curly medium-length hair, black square-framed sunglasses, short mustache and light beard, warm skin tone, relaxed half-smile. Do not inherit his grey t-shirt, food or restaurant background. WARDROBE (fixed): cream heavyweight boxy tee; open chocolate-brown suede overshirt, cuffs pushed up; cream wide-leg pleated wool trousers; cream-and-brown suede New Balance 9060 sneakers with a chunky layered sole; matte black square sunglasses never removed.
[Stage 1 — approx 0–5s] FASHION CUT. Six hard cuts of about 0.8s, no transitions, each a different lens, all on the plaza, white car in the far background. Cut 1: ground-level macro, a sneaker lands on concrete, trouser hem swings. Cut 2: 100mm close-up of the suede sleeve, hand pushing the cuff up. Cut 3: tight profile of the sunglasses, the white car reflected in the lens. Cut 4: 35mm handheld from behind, curls catching sun, car soft ahead. Cut 5: waist-level tracking of the trousers. Cut 6: high oblique drone at about 45 degrees, the man a small figure crossing toward the car, long shadow leading him. End state: man mid-plaza, walking left to right toward the driver door, car about 15 meters away. [Stage 2 — approx 5–11s] Continue from mid-plaza. Camera: one continuous drone shot, diagonal top-down at about 45 degrees, descending and drifting to hold man and car in one frame. He walks at a steady unhurried pace, hands relaxed. At 9s the drone is about 3 meters off the ground behind him as he reaches the driver door. End state: he stands at the driver door, right hand on the handle, sunglasses on. [Stage 3 — approx 11–18s] ASMR INTERIOR. Continue from hand on handle. Camera: restrained handheld, breath-level micro movement only, no gimbal smoothness. Shot A: close-up of the handle, the door unlatches and swings open. Shot B: from the passenger seat, he lowers into the cream seat, door pulls shut with one solid close. Shot C: macro of his hands settling on the caramel wheel, thumbs on the stitching. Shot D: macro of the seatbelt buckle clicking home. Shot E: macro of fingers on the ignition, one press at 17s; the dashboard wakes. End state: engine at idle, both hands on the wheel, sunglasses on, door closed. [Stage 4 — approx 18–26s] DRIVE. Continue from idle. Shot A: low static ground-level shot beside the rear wheel, the car pulls out of the plaza left to right, rear wing and red light bar crossing frame. Shot B: drone chase from a high oblique 45-degree angle behind, the car leaving the plaza onto a long empty two-lane asphalt road through pale sand terrain, the drone descending as the car accelerates; speed shown by road markings streaming past. Shot C: interior over-the-shoulder from the passenger headrest, his hand shifting, road unrolling through the windshield, sunlight sliding across the sunglasses. End state: car at speed on the empty road, no other vehicles. [Stage 5 — approx 26–30s] CLOSING. One continuous drone shot: from a rear-following position the drone rises and slows while the car continues away, becoming small on the straight road with its long shadow, the architecture faint on the horizon. Hold the final frame for the last second, car still moving. Final state: one white car receding on an empty road, no text.
No music at any point. Diegetic sound only, close-miked, ASMR clarity. Stage 1: <crisp sneaker steps on concrete, suede rustle, light dry wind, each cut lands on a footstep>. Stage 2: <footsteps growing closer, soft wind>. Stage 3: <door latch, leather creak as he sits, one deep solid door close, hands on stitched leather, seatbelt click, at 17s a single ignition press then a raw naturally aspirated flat-six cold start bark settling into a textured idle with low-frequency pressure>. Stage 4: <tires on asphalt, engine climbing the rev range, one clean upshift, exhaust echoing off the architecture then opening into wide space, wind rising>. Stage 5: <engine fading into distance, wind takes over, final second near silent>. No voiceover.
Exactly one man, the hero from @Image 2. Exactly one car, the white coupe from @Image 1. Keep face, curly hair, black square sunglasses, wardrobe, car color, white wheels, rear wing, red light bar and interior unchanged in every shot. Sunglasses stay on for all 30 seconds, including inside the car. Man walks left to right; car exits left to right, then away from camera. Sun low from the left throughout. No other people, vehicles or graphics.
no music, no voiceover, no subtitles, no on-screen text, no watermarks, no split-screen, no grid, no panel borders, no collage, no second car, no duplicate man, no face distortion, no identity change, no sunglasses removed, no wardrobe change, no grey t-shirt, no hat, no warped wheels, no morphing car body, no extra rear wings, no cartoon, no cheap CGI, no plastic skin, no flat lighting, no smooth gimbal in Stage 3, no night, no crowd.
CHECKPOINTS — FRAMES FROM THE GENERATED FILM


STEP 03
Anatomy — Why the Film Holds Together
- 1
The contract comes first
The FORMAT line is a genre pivot written as a spec: luxury fashion film into automotive ASMR drive, fast-cut opening, then long takes, no music. Seedance is told the rhythm will change before it is told what anything looks like — so the change reads as intent, not drift.
- 2
A sheet is a location, not a storyboard
The most important sentence in the prompt is a waiver: Ignore the grid layout. The panels are NOT keyframes, shot order or angles to copy. Without it, a six-panel reference tends to produce six-panel video. With it, the sheet becomes what it was built to be — a place the camera is free to move through.
- 3
One reference, one job
@Image 1 defines the car and the environment and nothing else. @Image 2 defines the hero's face and identity ONLY, and the prompt lists what it must not inherit — grey t-shirt, food, restaurant background. Every reference has a lane; nothing bleeds.
- 4
Wardrobe as a fixed variable
The WARDROBE (fixed) block dresses him piece by piece — boxy cream tee, chocolate suede overshirt with cuffs pushed up, pleated wool trousers, the layered-sole sneakers — and locks the sunglasses with never removed. CONSISTENCY repeats it: on for all 30 seconds, including inside the car. Say a lock twice and it holds.
- 5
End states hand the baton
Every stage closes with an End state — mid-plaza walking left to right, hand on the handle, engine at idle, at speed on the empty road — and the next stage opens with Continue from. That is scene lock stretched across cuts: the model always knows where the previous shot left the world.
- 6
Sound is the edit
With no music at any point, the AUDIO block has to carry structure. Each cut lands on a footstep turns the fashion cut into a rhythm; the ignition is pinned to a timestamp — one press at 17s — and answered by a naturally aspirated flat-six cold start bark. The soundtrack is a shot list you can hear.
- 7
The insurance
The NEGATIVE PROMPT mirrors every positive risk in the prompt: no extra rear wings for the identity lock, no grey t-shirt for the portrait waiver, no smooth gimbal in Stage 3 for the handheld ASMR, no split-screen, no grid, no collage for the sheet. Each ban is the shadow of a specific instruction.
THE RESULT
Output — The Cold Start

Check it against the prompt. The sunglasses never come off, inside or out. The car exits left to right and then straight away from camera, exactly as CONSISTENCY orders. The red light bar and the single swan-neck wing survive every angle, from the low static beside the rear wheel to the final drone that lets the car shrink into the road. One honest note for the audit: the FORMAT line asks for 30 seconds and the generation lands at about 22 — Seedance compresses long timelines, but all five stages are present and in order, which is what the end states were there to guarantee.

BUILD THE WORLD ONCE. THEN WALK INTO IT.
A reference sheet is a location scout.
The prompt is the shoot.
Swap the Porsche for a watch, a yacht, a pair of boots. Swap the plaza for a rooftop or a salt flat. The skeleton carries any luxury film: one sheet that defines the object and the place, one portrait that defines a face and nothing else, five stages that each hand over an end state, and a soundtrack made only of the things you can see.
Seedance 2.5 runs inside invideo — multi-reference inputs, timecoded prompts, one flat generation. Start with the link below.
WRITTEN, DIRECTED & PROMPTED BY DENIZ AKKABAK · @DENIZ.AKKABAK