Two brothers in a Lamborghini at night: Marcus, the younger, a famous rapper, drives, and Ray, the older, is not impressed — until you learn the whole conversation was in Marcus’s head. The car was never the point; the real test was how far AI video can go with acting.
The takes
Four takes from the final cut, in film order. Watch one with its sound, or beside its prompt. Copy prompt copies the whole prompt, exactly as it was sent.
The Settake 2.4
His wins, written as a friend’s camcorder tape from the crowd: six fragments in ten seconds. The film opens on the last one at 0:04 and cuts four more into the fame montage at 1:23.
Show prompt
SCENE CONTEXT a friend with a consumer VHS camcorder shoots Marcus's set from the wings and from inside the crowd of a low-ceilinged club packed past capacity, somewhere between a strip club and a sweatbox venue. what survives is six fragments hard-cut together, each shot from a different position at a different minute of the set, the tape jumping through the night. the camera is drowned in bodies, lifted over heads, shoved, and every fragment is fighting to keep Marcus in frame while the room detonates around him. the sound is a cheap camcorder microphone dying happily under club volume. FORMAT MODE ten seconds total, 4:3 full frame, interlaced consumer-camcorder cadence, six hard-cut fragments of roughly a second and a half each, no fragment exceeds two seconds, every cut a tape splice, no transitions. ACTIVE REFERENCES @Image1 -> Marcus: take his exact face geometry, skin, build and stage presence, rendered in monochrome. LOCATION MAP a low black-box club stage, knee height, no barrier. the crowd starts at the stage lip and packs back fifteen meters into darkness. a single follow spot hangs house-right, strobes on the back truss. the wings are a curtained slot at stage left where the camera starts. a speaker stack towers at the stage's right corner. CAST Marcus (@Image1), unrepeatable marker: he raps with his chin up, throwing lines down at the room like he is taller than the ceiling. everyone else on the tape is crowd: silhouettes, raised hands, one girl on shoulders as a dark shape against light. STAGING Marcus works the stage lip across all fragments, never static, lunging at the crowd and pulling back. the camera position changes every fragment: wings, crowd floor, over heads, crushed at the stage lip. bodies constantly cross and block the lens. WARDROBE Marcus in the white ribbed tank top now soaked dark at the chest, dog tags swinging, a towel jammed half into his back pocket. no change in any fragment. GAZE AND EYELINES Marcus plays the room, not the lens, in beats one, two, three and five. in beat four he finds the camera over the heads and points dead into it for half a second. in beat six his eyes are closed. the crowd never looks at the camera. EVENT AND SHARED DIRECTION a room of two thousand people and one man feeding on each other, and the tape barely keeping up. RULES monochrome throughout. hard cuts only. the camera never steadies and is never at eye-height twice in a row. every zoom is a degrading digital zoom. strobe and spot are allowed to blow the image to white and to black. the microphone distorts under bass on every fragment. BEAT 1 — from the wings. OPEN from the curtained slot at stage left, Marcus already mid-line in profile at mid-depth, spot carving him out of blackness, crowd a wall of hands beyond him. ACTION he lunges to the stage lip on a line -> the front row surges up to meet him -> he snaps back upright with the mic pressed to his mouth -> sweat flicks off him through the spotlight beam. CAMERA handheld micro-wobble from the wings, one slow drift forward. LENS MWS in profile, Marcus between the black curtain edge and the burning white crowd, ceiling pressing low. PERFORMANCE wants the room louder / blocked by a room already at its ceiling / plays a man winning an argument with two thousand people / turns on the front-row surge. DIALOGUE one bar of his verse, thrown down at the floor, the cadence merciless. {Told you I was him.} SFX <bass hits distorting the little microphone into crunch> EDIT hard cut on a strobe pop. BEAT 2 — inside the crowd. OPEN the frame is already half blocked by the back of a head, Marcus a burning figure above and beyond it, arms wide. ACTION the camera fights sideways for a gap -> hands shoot up through frame in rows like a wave -> Marcus drops to a crouch at the lip and the whole frame tilts trying to follow -> the crush shoves the operator and the world smears. CAMERA one shove-driven whip, uncontrolled, human. LENS MS from below through bodies, Marcus half-eclipsed, spotlight flaring across the lens in a soft vertical smear. PERFORMANCE wants to stay on him / blocked by two thousand shoulders / plays a fan first and a cameraman second / turns on the crouch, almost losing him. DIALOGUE none from him the mic can resolve, the room chanting one syllable in unison. SFX <a chant, two thousand voices, clipping to mud> EDIT hard cut mid-shove. BEAT 3 — the girl on shoulders. OPEN wide from the back of the room over a sea of heads, Marcus small and white-hot on stage at far depth, a girl rising onto shoulders in silhouette at mid-frame. ACTION she rises into the beam and becomes a black cutout against the spot -> her arms open in a slow V while the crowd bounces in double time around her -> Marcus points at her from the stage and the room roars -> the strobe fires and the whole image hammers white-black-white. CAMERA locked over-head hold, arms-up shot, no move. LENS EWS, 4:3, geometry of one bright man, one dark silhouette, a thousand hands. PERFORMANCE the room wants to be seen by him one person at a time / blocked by its own size / plays a single organism / turns on his pointing hand finding her. DIALOGUE none. SFX <the roar swallowing the music for a breath> EDIT hard cut inside the strobe. BEAT 4 — he finds the lens. OPEN crushed at the stage lip, shooting up at Marcus at full height, ceiling lights burning behind his head, the frame shaking with the crowd's bounce. ACTION his eyes sweep the front row -> they catch the camera and stop -> he leans down over the lip, closing half the distance, and points dead into the lens on the beat -> the grin opens faster than his eyes. CAMERA one digital crash zoom into his face, pixelating as it commits. LENS MS collapsing to MCU from below, his pointing hand huge and soft, his face hard behind it. PERFORMANCE wants the tape to know he saw it / blocked by a verse he never drops / plays generosity as dominance / turns on the point landing exactly on the downbeat. DIALOGUE half a bar, straight down the lens. {Yeah, you.} SFX <the front row screaming close to the mic> EDIT hard cut on the point. BEAT 5 — speaker stack. OPEN from beside the speaker stack at stage right, low angle across the stage, Marcus mid-frame in three-quarter back view, the crowd's faces catching spot-spill beyond him. ACTION he walks the lip house-left with the mic hand carving the air -> the towel finally falls from his pocket and he does not notice -> he plants at center and hits the last line of the verse with his whole back -> the crowd's hands come down on the beat like weather. CAMERA one slow pan following him left, nothing else. LENS MWS from behind and below, his silhouette rimmed by the follow spot, bass bins huge in foreground. PERFORMANCE wants to land the verse's last word like a verdict / blocked by nothing anymore / plays the best night of his life / turns on the hands coming down. DIALOGUE the last three words of the verse, roared. {Nobody like me!} SFX <the sub-bass rattling the camcorder's plastic body audibly> EDIT hard cut on the hands landing. BEAT 6 — eyes closed. OPEN tight on Marcus from the crowd, already mid-frame, chin up, eyes closed, mic lowered to his chest, drenched, the room's roar washing over him. ACTION he stands still for the first time on the tape -> the spot flares across his wet face and the auto-exposure hunts and blooms -> his chest rises once, huge -> the strobe fires and holds him in three frozen white instants. CAMERA locked, held over heads, trembling only with the operator's arms. LENS MCU from below, his face half-blown by the spot, blackness everywhere else. PERFORMANCE wants to keep this exact second forever / blocked by time / plays a man being loved at volume / turns on the single breath. DIALOGUE none. SFX <the roar, endless, clipping the microphone to a solid wall> EDIT end of fragment, tape stutter. PHYSICS sweat obeys gravity and flies on real arcs off a moving body. the crowd is a fluid mass: surges travel through it as waves, hands rise in rippling rows, the camera operator is shoved by real momentum and recovers late. the towel falls at true weight. the camcorder's plastic housing buzzes against the bass. LIGHTING one hard follow spot from house-right is the only real source, carving Marcus out and blowing his whites past clipping. back-truss strobes fire in short bursts that hammer the frame to full white then drop it to near black while the auto-exposure hunts. everything the spot does not touch lives as lifted, milky grey-black. no colored light survives the monochrome. AUDIO <club PA at crush volume, the camcorder mic distorting it to a warm wall> <crowd roar and chant, clipping> <bass thuds felt as microphone crunch> SCORE none, all music is the room's own PA, distorted and diegetic STYLE consumer VHS camcorder capture, interlaced, vertical smear off the follow spot, chroma stripped to full monochrome, lifted milky blacks, heavy tape grain, tracking noise flickering at the frame bottom on two fragments, auto-exposure hunting on every light change, digital crash zoom that pixelates as it commits. palette: lifted grey blacks 50, silver mids 30, clipped whites 20. TYPOGRAPHY none VOICE Marcus on the mic: young American rap delivery through a distorting PA, chest-deep, percussive consonants, total command, never rushed even at full volume. the crowd: an undifferentiated roar and one unison chant. POSITIVE LOCKS the image stays monochrome in every frame. every fragment boundary is a hard splice. Marcus's face matches @Image1 exactly wherever it resolves. the crowd remains silhouettes and dark shapes throughout; Marcus is the only resolved face on the tape. the camera height and position change on every fragment. the microphone stays cheap, overloaded and honest for the full ten seconds.
01Contextwritten once, true for the whole scene
SCENE CONTEXT a friend with a consumer VHS camcorder shoots Marcus's set from the wings and from inside the crowd of a low-ceilinged club packed past capacity, somewhere between a strip club and a sweatbox venue. what survives is six fragments hard-cut together, each shot from a different position at a different minute of the set, the tape jumping through the night. the camera is drowned in bodies, lifted over heads, shoved, and every fragment is fighting to keep Marcus in frame while the room detonates around him. the sound is a cheap camcorder microphone dying happily under club volume. FORMAT MODE ten seconds total, 4:3 full frame, interlaced consumer-camcorder cadence, six hard-cut fragments of roughly a second and a half each, no fragment exceeds two seconds, every cut a tape splice, no transitions. ACTIVE REFERENCES @Image1 -> Marcus: take his exact face geometry, skin, build and stage presence, rendered in monochrome. LOCATION MAP a low black-box club stage, knee height, no barrier. the crowd starts at the stage lip and packs back fifteen meters into darkness. a single follow spot hangs house-right, strobes on the back truss. the wings are a curtained slot at stage left where the camera starts. a speaker stack towers at the stage's right corner. CAST Marcus (@Image1), unrepeatable marker: he raps with his chin up, throwing lines down at the room like he is taller than the ceiling. everyone else on the tape is crowd: silhouettes, raised hands, one girl on shoulders as a dark shape against light. STAGING Marcus works the stage lip across all fragments, never static, lunging at the crowd and pulling back. the camera position changes every fragment: wings, crowd floor, over heads, crushed at the stage lip. bodies constantly cross and block the lens. WARDROBE Marcus in the white ribbed tank top now soaked dark at the chest, dog tags swinging, a towel jammed half into his back pocket. no change in any fragment. GAZE AND EYELINES Marcus plays the room, not the lens, in beats one, two, three and five. in beat four he finds the camera over the heads and points dead into it for half a second. in beat six his eyes are closed. the crowd never looks at the camera. EVENT AND SHARED DIRECTION a room of two thousand people and one man feeding on each other, and the tape barely keeping up. RULES monochrome throughout. hard cuts only. the camera never steadies and is never at eye-height twice in a row. every zoom is a degrading digital zoom. strobe and spot are allowed to blow the image to white and to black. the microphone distorts under bass on every fragment.
02Beatsone per shot, the part that changes
BEAT 1 — from the wings. OPEN from the curtained slot at stage left, Marcus already mid-line in profile at mid-depth, spot carving him out of blackness, crowd a wall of hands beyond him. ACTION he lunges to the stage lip on a line -> the front row surges up to meet him -> he snaps back upright with the mic pressed to his mouth -> sweat flicks off him through the spotlight beam. CAMERA handheld micro-wobble from the wings, one slow drift forward. LENS MWS in profile, Marcus between the black curtain edge and the burning white crowd, ceiling pressing low. PERFORMANCE wants the room louder / blocked by a room already at its ceiling / plays a man winning an argument with two thousand people / turns on the front-row surge. DIALOGUE one bar of his verse, thrown down at the floor, the cadence merciless. {Told you I was him.} SFX <bass hits distorting the little microphone into crunch> EDIT hard cut on a strobe pop. BEAT 2 — inside the crowd. OPEN the frame is already half blocked by the back of a head, Marcus a burning figure above and beyond it, arms wide. ACTION the camera fights sideways for a gap -> hands shoot up through frame in rows like a wave -> Marcus drops to a crouch at the lip and the whole frame tilts trying to follow -> the crush shoves the operator and the world smears. CAMERA one shove-driven whip, uncontrolled, human. LENS MS from below through bodies, Marcus half-eclipsed, spotlight flaring across the lens in a soft vertical smear. PERFORMANCE wants to stay on him / blocked by two thousand shoulders / plays a fan first and a cameraman second / turns on the crouch, almost losing him. DIALOGUE none from him the mic can resolve, the room chanting one syllable in unison. SFX <a chant, two thousand voices, clipping to mud> EDIT hard cut mid-shove. BEAT 3 — the girl on shoulders. OPEN wide from the back of the room over a sea of heads, Marcus small and white-hot on stage at far depth, a girl rising onto shoulders in silhouette at mid-frame. ACTION she rises into the beam and becomes a black cutout against the spot -> her arms open in a slow V while the crowd bounces in double time around her -> Marcus points at her from the stage and the room roars -> the strobe fires and the whole image hammers white-black-white. CAMERA locked over-head hold, arms-up shot, no move. LENS EWS, 4:3, geometry of one bright man, one dark silhouette, a thousand hands. PERFORMANCE the room wants to be seen by him one person at a time / blocked by its own size / plays a single organism / turns on his pointing hand finding her. DIALOGUE none. SFX <the roar swallowing the music for a breath> EDIT hard cut inside the strobe. BEAT 4 — he finds the lens. OPEN crushed at the stage lip, shooting up at Marcus at full height, ceiling lights burning behind his head, the frame shaking with the crowd's bounce. ACTION his eyes sweep the front row -> they catch the camera and stop -> he leans down over the lip, closing half the distance, and points dead into the lens on the beat -> the grin opens faster than his eyes. CAMERA one digital crash zoom into his face, pixelating as it commits. LENS MS collapsing to MCU from below, his pointing hand huge and soft, his face hard behind it. PERFORMANCE wants the tape to know he saw it / blocked by a verse he never drops / plays generosity as dominance / turns on the point landing exactly on the downbeat. DIALOGUE half a bar, straight down the lens. {Yeah, you.} SFX <the front row screaming close to the mic> EDIT hard cut on the point. BEAT 5 — speaker stack. OPEN from beside the speaker stack at stage right, low angle across the stage, Marcus mid-frame in three-quarter back view, the crowd's faces catching spot-spill beyond him. ACTION he walks the lip house-left with the mic hand carving the air -> the towel finally falls from his pocket and he does not notice -> he plants at center and hits the last line of the verse with his whole back -> the crowd's hands come down on the beat like weather. CAMERA one slow pan following him left, nothing else. LENS MWS from behind and below, his silhouette rimmed by the follow spot, bass bins huge in foreground. PERFORMANCE wants to land the verse's last word like a verdict / blocked by nothing anymore / plays the best night of his life / turns on the hands coming down. DIALOGUE the last three words of the verse, roared. {Nobody like me!} SFX <the sub-bass rattling the camcorder's plastic body audibly> EDIT hard cut on the hands landing. BEAT 6 — eyes closed. OPEN tight on Marcus from the crowd, already mid-frame, chin up, eyes closed, mic lowered to his chest, drenched, the room's roar washing over him. ACTION he stands still for the first time on the tape -> the spot flares across his wet face and the auto-exposure hunts and blooms -> his chest rises once, huge -> the strobe fires and holds him in three frozen white instants. CAMERA locked, held over heads, trembling only with the operator's arms. LENS MCU from below, his face half-blown by the spot, blackness everywhere else. PERFORMANCE wants to keep this exact second forever / blocked by time / plays a man being loved at volume / turns on the single breath. DIALOGUE none. SFX <the roar, endless, clipping the microphone to a solid wall> EDIT end of fragment, tape stutter.
03Prompt segmentsthe saved look, sound and style, reused across prompts
PHYSICS sweat obeys gravity and flies on real arcs off a moving body. the crowd is a fluid mass: surges travel through it as waves, hands rise in rippling rows, the camera operator is shoved by real momentum and recovers late. the towel falls at true weight. the camcorder's plastic housing buzzes against the bass. LIGHTING one hard follow spot from house-right is the only real source, carving Marcus out and blowing his whites past clipping. back-truss strobes fire in short bursts that hammer the frame to full white then drop it to near black while the auto-exposure hunts. everything the spot does not touch lives as lifted, milky grey-black. no colored light survives the monochrome. AUDIO <club PA at crush volume, the camcorder mic distorting it to a warm wall> <crowd roar and chant, clipping> <bass thuds felt as microphone crunch> SCORE none, all music is the room's own PA, distorted and diegetic STYLE consumer VHS camcorder capture, interlaced, vertical smear off the follow spot, chroma stripped to full monochrome, lifted milky blacks, heavy tape grain, tracking noise flickering at the frame bottom on two fragments, auto-exposure hunting on every light change, digital crash zoom that pixelates as it commits. palette: lifted grey blacks 50, silver mids 30, clipped whites 20. TYPOGRAPHY none VOICE Marcus on the mic: young American rap delivery through a distorting PA, chest-deep, percussive consonants, total command, never rushed even at full volume. the crowd: an undifferentiated roar and one unison chant. POSITIVE LOCKS the image stays monochrome in every frame. every fragment boundary is a hard splice. Marcus's face matches @Image1 exactly wherever it resolves. the crowd remains silhouettes and dark shapes throughout; Marcus is the only resolved face on the tape. the camera height and position change on every fragment. the microphone stays cheap, overloaded and honest for the full ten seconds.
The Drivetake 65.24
Part 2 of the car conversation, one 30-second prompt from “I send money.” to the second “Marcus.” In the film at 0:19–0:47.
Show prompt
SCENE CONTEXT: the same two brothers, the same matte blue-grey Lamborghini, further into the same downtown night drive. The small talk is over and the older one has started asking questions the younger one has no prepared answers for. The exchange still runs at the volume of two men who have run out of ways to talk, and the car never stops moving. FORMAT MODE: 30 seconds, 21:9, 24fps, four beats, three hard cuts, synchronized audio on. ACTIVE REFERENCES: @Image1 -> MARCUS: face, skin texture, shoulder-length dark locs with faded temples, the diamond grill, the silver nose ring, the ear studs, the neck and chest fine-line tattoos, the white pearls over the heavy iced chain, the teal-navy-cream-rust paisley patchwork track jacket. @Image2 -> RAY: face, skin texture, the close-shaved head, the pale scar splitting the right eyebrow, the navy polo shirt. @Image3 -> THE CAR: exact body shape, matte blue-grey paint, wheels, light signature — every visible piece of the car inside and out belongs to this exact vehicle. @Image4 -> FRAMING PLATE ONLY: the widest permitted frame — a cross-cabin medium of the driver from the passenger side; take framing, camera height and light temperature only, NOT the parked background, this car is moving. @Image5 -> FRAMING PLATE ONLY: the matching cross-cabin medium of the passenger from the driver's side; same rule, framing and light only. LOCATION MAP: a left-hand-drive American car. The steering wheel is mounted on the car's LEFT side. MARCUS sits in the LEFT front seat driving, both hands on the wheel. RAY sits in the RIGHT front passenger seat. A wide carbon-fibre centre tunnel runs between them like a wall. Through the glass a multi-lane downtown avenue streams past between glass towers, streetlights and signage passing at every depth, continuously, in every shot. CAST: MARCUS, a Black man in his late twenties, lean, mid-brown skin with visible pore texture, shoulder-length dark locs hanging forward, temples faded short, a full diamond grill across his upper teeth, a small silver ring in his left nostril, a diamond stud in each ear, fine-line tattoos across neck and upper chest, white pearls over a heavy iced chain. RAY, mid thirties, darker and heavier, close-shaved head, a pale scar splitting the right eyebrow — the SAME jaw and the SAME eyes, so they read as blood at a glance. STAGING: both stay seated throughout. MARCUS keeps both hands on the wheel. RAY turns fully toward him exactly once, on beat three, and does NOT turn back. Every camera position in this clip is INSIDE the cabin, at head height, mounted at the doors facing across or forward — never behind the seats, never showing the dashboard from the rear cabin, never outside the glass. The widest frame in the whole clip is the cross-cabin medium of the framing plates; every other frame is tighter. WARDROBE: MARCUS in the zip-front patchwork bandana-paisley track jacket panelled in teal, navy, cream and rust-red, iced silver bracelet on the right wrist, pearls over chain, grill in. RAY in a plain navy polo shirt, no jewellery of any kind. EXACT in every beat, identical to @Image1 and @Image2. GAZE AND EYELINES: a ladder. Beats one and two, neither looks at the other. Beat three, RAY turns and looks straight at MARCUS and holds it; MARCUS meets his eyes ONCE, for well under a second, on 'Everybody say that.' — then the road pulls his eyes back and they stay there. Beat four, MARCUS's eyes never leave the windscreen while his brother says his name twice. NOBODY looks at camera. EVENT AND SHARED DIRECTION: the event, containing both men and never spoken — one brother refuses to be paid off, and the other runs out of currencies. MARCUS pursues it through the wheel and the road; RAY through stillness and questions nobody has asked him in years. THE SHARED DIRECTION: keep it ordinary. Do NOT let this become a scene. The essential axis: performed ease against unperformed presence. RULES: night-for-night throughout. The car is moving in EVERY frame — city lights visibly stream past the glass in every single shot and the car NEVER stops, slows to a halt, or waits at a light. Real-time, never slow motion. The SAME two men in the SAME clothes in every beat. Every camera inside the cabin, front or side, never behind. BEAT 1 — I send money (0-8s) OPEN: tight close on MARCUS from the passenger side, framed inside the width of @Image4, both hands on the wheel, a passing light arriving across his face, already mid-answer. The city streams behind the glass. CAMERA: locked off. The frame does NOT travel. LENS: close-up, the near eye sharpest, RAY out of frame entirely. ACTION: MARCUS answers fast and prepared -> RAY corrects him from off-screen without leaning on it -> MARCUS does not answer that -> the shot holds on him three more seconds while the correction sits there and the lights keep crossing his face. PERFORMANCE: MARCUS wants the ledger to count; blocked by it not counting; plays certainty; turns on the correction landing without any anger to push against. RAY off-screen wants one true answer; plays it quieter than the question deserved. DIALOGUE: fast, prepared: {I send money.} RAY, off-screen, quieter, not leaning on it: {That ain't what she asked.} Keep 'ain't' exactly — the dialect is locked and must not be corrected. SFX: <tyre roar> <a road seam arriving through both seats> EDIT: hard cut. BEAT 2 — what you eat today (8-16s) OPEN: tight close on RAY from the driver's side, framed inside the width of @Image5, lit by what passes outside and the cluster spill, already looking straight ahead, navy polo reading clearly. CAMERA: locked off. LENS: close-up matched to the previous beat, MARCUS out of frame. ACTION: RAY asks the smallest question in the film -> nothing comes back -> three full seconds of nothing while he breathes and blinks at his own cadence -> he lets it go, half-voiced, almost to himself, eyes never leaving the windscreen. PERFORMANCE: RAY wants any true answer of any size; blocked by rehearsed ones; plays it the way you check a fuel gauge — NOT tender, NOT concerned; turns on the silence being the answer. DIALOGUE: the way you check a fuel gauge: {What you eat today?} Then, after the silence, falling, a period not a question: {Right.} The word is 'Right.' spoken as a flat statement with falling pitch — one word, closing a subject, never rising, never 'all right'. SFX: <tyre roar> <one breath in and not out> EDIT: hard cut. BEAT 3 — I ain't everybody (16-25s) OPEN: an in-cabin over-shoulder from the passenger door — the camera physically mounted at the passenger door beside RAY's near shoulder, his shoulder and profile soft in the right foreground, MARCUS sharp beyond him at the wheel, both men in frame, no dashboard-from-behind, no rear cabin. Already moving, lights crossing both faces. CAMERA: one hand-held jitter-in-place, breathing with the volley, travelling nowhere. LENS: medium over-shoulder two-shot — the only two-shot in the clip, and no wider than the framing plates. ACTION: MARCUS reaches for the old deflection and hears it fail as it leaves his mouth, the grill catching a passing light exactly once on that line and never again -> RAY answers flat -> MARCUS pushes it one word further -> RAY turns fully toward him and does not turn back -> MARCUS meets his eyes once, under a second, on his last line, then the road pulls his eyes back. PERFORMANCE: MARCUS wants the old currency to still spend; blocked by his brother not taking it; plays the deflection and hears it die; turns on the eye contact costing more than the words. RAY wants him to hear his own sentence; blocked by nothing now — he has stopped waiting; plays stillness, then the turn; turns on deciding the questions are over. DIALOGUE: MARCUS, the deflection arriving already tired: {Everybody want somethin'.} RAY, flat: {I don't.} MARCUS, one word further, and hearing it fail, with the single brief meeting of eyes: {Everybody say that.} RAY, turned toward him now, even, final: {I ain't everybody.} Keep every line exactly as written — 'Everybody want somethin'' with no g and no 's' on want, and 'I ain't everybody' — the dialect is locked and must not be corrected to standard grammar. SFX: <tyre roar> <the indicator ticking once and stopping> EDIT: hard cut. BEAT 4 — the name (25-30s) OPEN: tight close on MARCUS from the passenger side, tighter than beat one, eyes forward, jaw set, hands on the wheel, lights crossing his face at driving rhythm. CAMERA: locked off. LENS: close-up, the near eye sharpest, RAY out of frame. ACTION: RAY says the name from off-screen -> MARCUS's jaw sets and his blink rate rises but his eyes never leave the road -> MARCUS answers with the only word he has left -> RAY says the name again, softer, and nothing comes back -> hold on MARCUS to the end, mouth closed, the grill never catching light again. PERFORMANCE: MARCUS wants the name unsaid; blocked by it already being said twice; plays stone with a rising cost visible only in the eyes; turns on the second saying. RAY wants his brother back for one second; plays the name instead of any question. DIALOGUE: RAY, off-screen, level: {Marcus.} MARCUS, low, not a warning, a request: {Don't.} RAY, off-screen, softer: {Marcus.} SFX: <tyre roar steady> <the cabin's own quiet> EDIT: hold to end. PHYSICS: the car is a low two-seat supercar with real mass — road seams arrive through both seats as one soft jolt, bodies sway millimetres on lane changes, the pearls and chain shift with the motion, the wheel makes constant small corrections under MARCUS's hands. Outside light sources travel past at consistent urban driving speed in every shot. LIGHTING: sodium and LED streetlights pass overhead as travelling warm streaks across faces; cool blue-green signage fills from the sides; the instrument cluster throws a low warm spill up onto both men; skin warm against the cold city, shadows open, never crushed. AUDIO: <steady tyre roar and low engine note under everything> <passing cars arriving and going> <insulated cabin quiet between lines> SCORE: none. No music of any kind for the entire clip. STYLE: photoreal night-for-night cinema, ARRI ALEXA aesthetic, 50mm and 85mm depth, shallow focus, horizontal-ellipse bokeh on the city lights, palette warm-skin 45 / teal-city 35 / neutral-black 20. TYPOGRAPHY: none. VOICE: both men Black American, unforced natural cadence, low volume, the register of a private car at night. MARCUS: late-twenties, lighter, quicker. RAY: mid-thirties, lower, slower, even. Every line spoken exactly as written in the braces, word for word, including 'That ain't what she asked', 'Everybody want somethin'', and 'I ain't everybody' — the dialect is the character and must not be corrected. 'Right.' falls, it never rises. Neither man raises his voice. POSITIVE LOCKS: the steering wheel is on the car's left in every frame. MARCUS drives from the left seat in every frame. The car moves in every frame with city light streaming past the glass. Every camera position is inside the cabin, at the doors or between the seats, facing across or forward. The widest frame equals the cross-cabin medium of the framing plates; every other frame is tighter. The grill catches light exactly once, on 'Everybody say that.' beat's deflection line, and never again. Faces stay asymmetric, blinking naturally, eyes wet with real catchlights, chests visibly breathing.
01Contextwritten once, true for the whole scene
SCENE CONTEXT: the same two brothers, the same matte blue-grey Lamborghini, further into the same downtown night drive. The small talk is over and the older one has started asking questions the younger one has no prepared answers for. The exchange still runs at the volume of two men who have run out of ways to talk, and the car never stops moving. FORMAT MODE: 30 seconds, 21:9, 24fps, four beats, three hard cuts, synchronized audio on. ACTIVE REFERENCES: @Image1 -> MARCUS: face, skin texture, shoulder-length dark locs with faded temples, the diamond grill, the silver nose ring, the ear studs, the neck and chest fine-line tattoos, the white pearls over the heavy iced chain, the teal-navy-cream-rust paisley patchwork track jacket. @Image2 -> RAY: face, skin texture, the close-shaved head, the pale scar splitting the right eyebrow, the navy polo shirt. @Image3 -> THE CAR: exact body shape, matte blue-grey paint, wheels, light signature — every visible piece of the car inside and out belongs to this exact vehicle. @Image4 -> FRAMING PLATE ONLY: the widest permitted frame — a cross-cabin medium of the driver from the passenger side; take framing, camera height and light temperature only, NOT the parked background, this car is moving. @Image5 -> FRAMING PLATE ONLY: the matching cross-cabin medium of the passenger from the driver's side; same rule, framing and light only. LOCATION MAP: a left-hand-drive American car. The steering wheel is mounted on the car's LEFT side. MARCUS sits in the LEFT front seat driving, both hands on the wheel. RAY sits in the RIGHT front passenger seat. A wide carbon-fibre centre tunnel runs between them like a wall. Through the glass a multi-lane downtown avenue streams past between glass towers, streetlights and signage passing at every depth, continuously, in every shot. CAST: MARCUS, a Black man in his late twenties, lean, mid-brown skin with visible pore texture, shoulder-length dark locs hanging forward, temples faded short, a full diamond grill across his upper teeth, a small silver ring in his left nostril, a diamond stud in each ear, fine-line tattoos across neck and upper chest, white pearls over a heavy iced chain. RAY, mid thirties, darker and heavier, close-shaved head, a pale scar splitting the right eyebrow — the SAME jaw and the SAME eyes, so they read as blood at a glance. STAGING: both stay seated throughout. MARCUS keeps both hands on the wheel. RAY turns fully toward him exactly once, on beat three, and does NOT turn back. Every camera position in this clip is INSIDE the cabin, at head height, mounted at the doors facing across or forward — never behind the seats, never showing the dashboard from the rear cabin, never outside the glass. The widest frame in the whole clip is the cross-cabin medium of the framing plates; every other frame is tighter. WARDROBE: MARCUS in the zip-front patchwork bandana-paisley track jacket panelled in teal, navy, cream and rust-red, iced silver bracelet on the right wrist, pearls over chain, grill in. RAY in a plain navy polo shirt, no jewellery of any kind. EXACT in every beat, identical to @Image1 and @Image2. GAZE AND EYELINES: a ladder. Beats one and two, neither looks at the other. Beat three, RAY turns and looks straight at MARCUS and holds it; MARCUS meets his eyes ONCE, for well under a second, on 'Everybody say that.' — then the road pulls his eyes back and they stay there. Beat four, MARCUS's eyes never leave the windscreen while his brother says his name twice. NOBODY looks at camera. EVENT AND SHARED DIRECTION: the event, containing both men and never spoken — one brother refuses to be paid off, and the other runs out of currencies. MARCUS pursues it through the wheel and the road; RAY through stillness and questions nobody has asked him in years. THE SHARED DIRECTION: keep it ordinary. Do NOT let this become a scene. The essential axis: performed ease against unperformed presence. RULES: night-for-night throughout. The car is moving in EVERY frame — city lights visibly stream past the glass in every single shot and the car NEVER stops, slows to a halt, or waits at a light. Real-time, never slow motion. The SAME two men in the SAME clothes in every beat. Every camera inside the cabin, front or side, never behind.
02Beatsone per shot, the part that changes
BEAT 1 — I send money (0-8s) OPEN: tight close on MARCUS from the passenger side, framed inside the width of @Image4, both hands on the wheel, a passing light arriving across his face, already mid-answer. The city streams behind the glass. CAMERA: locked off. The frame does NOT travel. LENS: close-up, the near eye sharpest, RAY out of frame entirely. ACTION: MARCUS answers fast and prepared -> RAY corrects him from off-screen without leaning on it -> MARCUS does not answer that -> the shot holds on him three more seconds while the correction sits there and the lights keep crossing his face. PERFORMANCE: MARCUS wants the ledger to count; blocked by it not counting; plays certainty; turns on the correction landing without any anger to push against. RAY off-screen wants one true answer; plays it quieter than the question deserved. DIALOGUE: fast, prepared: {I send money.} RAY, off-screen, quieter, not leaning on it: {That ain't what she asked.} Keep 'ain't' exactly — the dialect is locked and must not be corrected. SFX: <tyre roar> <a road seam arriving through both seats> EDIT: hard cut. BEAT 2 — what you eat today (8-16s) OPEN: tight close on RAY from the driver's side, framed inside the width of @Image5, lit by what passes outside and the cluster spill, already looking straight ahead, navy polo reading clearly. CAMERA: locked off. LENS: close-up matched to the previous beat, MARCUS out of frame. ACTION: RAY asks the smallest question in the film -> nothing comes back -> three full seconds of nothing while he breathes and blinks at his own cadence -> he lets it go, half-voiced, almost to himself, eyes never leaving the windscreen. PERFORMANCE: RAY wants any true answer of any size; blocked by rehearsed ones; plays it the way you check a fuel gauge — NOT tender, NOT concerned; turns on the silence being the answer. DIALOGUE: the way you check a fuel gauge: {What you eat today?} Then, after the silence, falling, a period not a question: {Right.} The word is 'Right.' spoken as a flat statement with falling pitch — one word, closing a subject, never rising, never 'all right'. SFX: <tyre roar> <one breath in and not out> EDIT: hard cut. BEAT 3 — I ain't everybody (16-25s) OPEN: an in-cabin over-shoulder from the passenger door — the camera physically mounted at the passenger door beside RAY's near shoulder, his shoulder and profile soft in the right foreground, MARCUS sharp beyond him at the wheel, both men in frame, no dashboard-from-behind, no rear cabin. Already moving, lights crossing both faces. CAMERA: one hand-held jitter-in-place, breathing with the volley, travelling nowhere. LENS: medium over-shoulder two-shot — the only two-shot in the clip, and no wider than the framing plates. ACTION: MARCUS reaches for the old deflection and hears it fail as it leaves his mouth, the grill catching a passing light exactly once on that line and never again -> RAY answers flat -> MARCUS pushes it one word further -> RAY turns fully toward him and does not turn back -> MARCUS meets his eyes once, under a second, on his last line, then the road pulls his eyes back. PERFORMANCE: MARCUS wants the old currency to still spend; blocked by his brother not taking it; plays the deflection and hears it die; turns on the eye contact costing more than the words. RAY wants him to hear his own sentence; blocked by nothing now — he has stopped waiting; plays stillness, then the turn; turns on deciding the questions are over. DIALOGUE: MARCUS, the deflection arriving already tired: {Everybody want somethin'.} RAY, flat: {I don't.} MARCUS, one word further, and hearing it fail, with the single brief meeting of eyes: {Everybody say that.} RAY, turned toward him now, even, final: {I ain't everybody.} Keep every line exactly as written — 'Everybody want somethin'' with no g and no 's' on want, and 'I ain't everybody' — the dialect is locked and must not be corrected to standard grammar. SFX: <tyre roar> <the indicator ticking once and stopping> EDIT: hard cut. BEAT 4 — the name (25-30s) OPEN: tight close on MARCUS from the passenger side, tighter than beat one, eyes forward, jaw set, hands on the wheel, lights crossing his face at driving rhythm. CAMERA: locked off. LENS: close-up, the near eye sharpest, RAY out of frame. ACTION: RAY says the name from off-screen -> MARCUS's jaw sets and his blink rate rises but his eyes never leave the road -> MARCUS answers with the only word he has left -> RAY says the name again, softer, and nothing comes back -> hold on MARCUS to the end, mouth closed, the grill never catching light again. PERFORMANCE: MARCUS wants the name unsaid; blocked by it already being said twice; plays stone with a rising cost visible only in the eyes; turns on the second saying. RAY wants his brother back for one second; plays the name instead of any question. DIALOGUE: RAY, off-screen, level: {Marcus.} MARCUS, low, not a warning, a request: {Don't.} RAY, off-screen, softer: {Marcus.} SFX: <tyre roar steady> <the cabin's own quiet> EDIT: hold to end.
03Prompt segmentsthe saved look, sound and style, reused across prompts
PHYSICS: the car is a low two-seat supercar with real mass — road seams arrive through both seats as one soft jolt, bodies sway millimetres on lane changes, the pearls and chain shift with the motion, the wheel makes constant small corrections under MARCUS's hands. Outside light sources travel past at consistent urban driving speed in every shot. LIGHTING: sodium and LED streetlights pass overhead as travelling warm streaks across faces; cool blue-green signage fills from the sides; the instrument cluster throws a low warm spill up onto both men; skin warm against the cold city, shadows open, never crushed. AUDIO: <steady tyre roar and low engine note under everything> <passing cars arriving and going> <insulated cabin quiet between lines> SCORE: none. No music of any kind for the entire clip. STYLE: photoreal night-for-night cinema, ARRI ALEXA aesthetic, 50mm and 85mm depth, shallow focus, horizontal-ellipse bokeh on the city lights, palette warm-skin 45 / teal-city 35 / neutral-black 20. TYPOGRAPHY: none. VOICE: both men Black American, unforced natural cadence, low volume, the register of a private car at night. MARCUS: late-twenties, lighter, quicker. RAY: mid-thirties, lower, slower, even. Every line spoken exactly as written in the braces, word for word, including 'That ain't what she asked', 'Everybody want somethin'', and 'I ain't everybody' — the dialect is the character and must not be corrected. 'Right.' falls, it never rises. Neither man raises his voice. POSITIVE LOCKS: the steering wheel is on the car's left in every frame. MARCUS drives from the left seat in every frame. The car moves in every frame with city light streaming past the glass. Every camera position is inside the cabin, at the doors or between the seats, facing across or forward. The widest frame equals the cross-cabin medium of the framing plates; every other frame is tighter. The grill catches light exactly once, on 'Everybody say that.' beat's deflection line, and never again. Faces stay asymmetric, blinking naturally, eyes wet with real catchlights, chests visibly breathing.
The Kitchentake 2.13
A flashback: the night his mother made Marcus, at sixteen, hear one sentence. Its last four fragments play at 1:07–1:11.
Show prompt
SCENE CONTEXT a memory from the same small apartment as @Video1, but a gentler evening than the fight: the night the Mother crossed the kitchen and made her son hear one sentence. Marcus at sixteen has come home carrying something that was done to him or near him, and she knows the shape of it without being told. the camera is the same witness that cannot hold still, a super wide lens inches from hands and faces, and the mother exists exactly as she always has: hands, a housedress hem, a shoulder and hair edge, a voice, never a whole figure and never a resolved face. six hard-cut fragments, each under two seconds, out of order the way tenderness gets filed next to grief. the same late golden light through thin curtains carries what the words cannot. FORMAT MODE ten seconds total, 2.39 widescreen, 35mm color film cadence, six hard-cut fragments of roughly a second and a half each, no fragment exceeds two seconds, every cut a splice, no transitions. ACTIVE REFERENCES @Image1 -> Marcus at sixteen: take his exact face geometry, skin, build, braids and slight frame, in natural color. @Video1 -> the same kitchen and hallway, the same golden window light and color register, the Mother's voice timbre and the hard-cut fragment rhythm. LOCATION MAP the narrow kitchen from @Video1: small table with two chairs against the wall, the window with thin curtains above the sink burning with low evening sun, the hallway of three doors opening off the kitchen, Marcus's bedroom door at its end. CAST Marcus at 16 (@Image1), unrepeatable marker: he receives comfort with his chin down and his eyes up, the same way he argues. the Mother, present only as hands, forearms, a housedress hem, bare feet, a shoulder and hair edge, and a voice that tonight does not crack because she will not allow it. she is never framed whole and her face never resolves. STAGING the wide lens lives inside the embrace-distance, half a meter or closer, so hands and faces bend at the frame edges. across the fragments she advances and he does not retreat, the reverse of every fight. counted exception: in beat six neither of them moves at all. WARDROBE Marcus in the washed teal hoodie over a white tee, grey jeans. the Mother's housedress is the pale florals gone grey with washing from @Video1. no changes within any fragment. GAZE AND EYELINES Marcus looks at her hands in beats one and three, at the floor in beat two, and in beat five his eyes finally lift to where her face would be, just above the frame's edge, and stay there. he never looks at the lens. the camera looks where a son looks: at hands that raised him. EVENT AND SHARED DIRECTION a mother installing one sentence in her son deep enough that it will still be there when he is famous. RULES natural color throughout, 35mm film texture matching @Video1 exactly. hard cuts only. the camera never steadies and never rises above chest height. the mother appears only in parts. every light source renders as a soft, warm, contained glow within its own source. the lens stays super wide and close. BEAT 1 — her hands arriving. OPEN super wide inches from Marcus's chest, his hands hanging useless at his sides at frame right, her two hands already entering from frame left and closing around one of his. ACTION her thumbs press across his knuckles, an old habit -> his fingers stay slack a moment, then curl -> the golden window light lies across both their hands like a third hand -> her grip tightens once, deliberate. CAMERA handheld micro-wobble, drifting a palm's width down toward the hands, one move. LENS ECU on hands warping wide at the edges, the kitchen soft gold beyond. PERFORMANCE she wants his hands to stop being fists / blocked by a boy who came home wearing the street / plays certainty like a tool she has used before / turns on his fingers curling. DIALOGUE none. SFX <the kitchen clock, the refrigerator hum> EDIT hard cut on the grip. BEAT 2 — his face, not ready. OPEN Marcus's face fills the super wide frame from centimeters, chin down, eyes down, jaw tight, one cheek gold and one in shadow, a shine at the lower lids he is managing. ACTION his jaw works once on nothing -> his breath drags in through his nose, long -> his eyes close and squeeze and open again wetter -> her shadow crosses the gold on his cheek as she steps nearer. CAMERA locked in trembling hands, no move, the shake is pulse. LENS ECU super wide, his forehead and chin bending at the frame edge. PERFORMANCE wants to say it was not like they said / blocked by a throat that has closed for business / plays composed and is not / turns on her shadow arriving. DIALOGUE none. SFX <a floorboard's slow creak as her weight arrives> EDIT hard cut on the shadow. BEAT 3 — her hands take his face. OPEN both her hands entering the super wide frame from below and left, rising to hold Marcus's jaw on both sides, her wrists and forearms and the frame's edge cutting her at the shoulders, her face above the top of frame, never shown. ACTION her palms settle along his jaw and his whole head lowers a centimeter into them -> her thumb wipes the cheekbone once, taking what was there -> his eyes stay down and she waits -> her hands lift his face one degree, insisting. CAMERA handheld micro-wobble, holding, no move. LENS MCU super wide, boy and hands, her body an out-of-focus warmth at frame left. PERFORMANCE she wants his eyes / blocked by shame that is not his but sits on him anyway / plays patience like iron / turns on the thumb wipe. DIALOGUE her voice above the frame, low, level, each word placed. {Marcus. It's not your fault.} SFX <the pot's low simmer from the stove> EDIT hard cut on the last word. BEAT 4 — the hem, closer than ever. OPEN low at knee height, super wide: her housedress hem and bare feet at frame center, closer to his sneakers than the two of them have stood since the fight, the hallway's long gold shadows behind. ACTION her feet take one final small step until the hem brushes his shin -> his sneakers hold their ground, toes flexing once inside the canvas -> the hem sways and stills -> both their shadows on the floorboards merge into one shape. CAMERA handheld at knee height, sinking one breath lower, one move. LENS WS from the floor, bodies as fragments, the corridor bending away super wide. PERFORMANCE she wants zero distance / blocked by nothing tonight, and that is the event / plays proximity as the whole argument / turns on the shadows merging. DIALOGUE none. SFX <fabric brushing denim, the clock further away now> EDIT hard cut on the merged shadow. BEAT 5 — you hear me. OPEN Marcus's face again, super wide and inches away, but lifted now, eyes finally up to where her face lives just above the frame's top edge, the gold full on him, wet tracks shining and permitted. ACTION his eyes hold on her unseen face for the first time in the memory -> her hand slides from his jaw to the back of his neck -> his chin fights her sentence and loses -> he nods once, small, real, the nod of a contract. CAMERA locked, trembling only with pulse, no move. LENS ECU super wide, his lifted face bending gently at the edges. PERFORMANCE she wants the sentence signed for / blocked by a boy who believes blame keeps him safe / plays the question that is not a question / turns on the nod. DIALOGUE her voice above the frame, harder now, love with its sleeves rolled. {You hear me!} SFX <her hand settling on his neck, cloth-soft> EDIT hard cut on the nod. BEAT 6 — held. OPEN the kitchen from the hallway floor, super wide and low and finally still: the two of them at the window as one shape, her arms around him, his face buried at her shoulder, her head bent to his, both faces unresolved against the burning gold curtains. ACTION steam rises off the pot nobody is watching -> dust turns in the window shaft around them -> his hand comes up her back and grips the grey floral cotton -> neither of them moves again. CAMERA locked, resting on the floor, breath-still. LENS WS deep frame, one shape made of two people, the light the only witness with a whole view. PERFORMANCE the room wants this to be the memory that wins / blocked by all the other memories / plays the silence after the best sentence / turns on his hand gripping her dress. DIALOGUE none. SFX <the pot's simmer, the clock, the street faint through glass> EDIT hold a beat, end. PHYSICS her thumb on wet skin drags and clears at true friction. the hem sways at cotton weight and settles. his nod carries the small mass of a real head, no more. steam and window dust obey slow indoor air. two bodies leaning into each other carry real weight and settle rather than pose. LIGHTING the same single low evening sun through thin kitchen curtains as @Video1, gold, raking, blooming softly where it strikes, handing the hallway to warm shadow. every light source renders as a soft, warm, contained glow within its own source; faces carry natural skin texture, pores, oil and tear-shine, no gloss. AUDIO <small-apartment room tone, clock, refrigerator hum> <a pot simmering> <street sounds faint through glass> SCORE none STYLE 35mm color film capture, Kodak palette, visible grain, halation on the window highlights, super wide angle lens inches from subjects so the room bends at the edges, handheld throughout except the final locked fragment, natural sources only, memory-real not staged. palette: warm gold 40, deep amber shadow 35, washed teal accents 25. TYPOGRAPHY none VOICE the Mother: American, low and level, the register of a woman who prays and works doubles, tonight refusing to crack because the sentence needs a steady floor; the exclamation lands as pressure, not volume. Marcus at 16: says nothing; his breath and one nod are his whole part. POSITIVE LOCKS the image stays natural 35mm color in every frame and matches @Video1's register exactly. every fragment boundary is a hard splice. Marcus's face matches @Image1 exactly wherever it resolves. the Mother appears only as hands, forearms, hem, bare feet, shoulder and hair edge, and her face never resolves in any frame; her dialogue arrives from above or beside the frame. the lens stays super wide and within arm's reach. the camera never rises above chest height for the full ten seconds.
01Contextwritten once, true for the whole scene
SCENE CONTEXT a memory from the same small apartment as @Video1, but a gentler evening than the fight: the night the Mother crossed the kitchen and made her son hear one sentence. Marcus at sixteen has come home carrying something that was done to him or near him, and she knows the shape of it without being told. the camera is the same witness that cannot hold still, a super wide lens inches from hands and faces, and the mother exists exactly as she always has: hands, a housedress hem, a shoulder and hair edge, a voice, never a whole figure and never a resolved face. six hard-cut fragments, each under two seconds, out of order the way tenderness gets filed next to grief. the same late golden light through thin curtains carries what the words cannot. FORMAT MODE ten seconds total, 2.39 widescreen, 35mm color film cadence, six hard-cut fragments of roughly a second and a half each, no fragment exceeds two seconds, every cut a splice, no transitions. ACTIVE REFERENCES @Image1 -> Marcus at sixteen: take his exact face geometry, skin, build, braids and slight frame, in natural color. @Video1 -> the same kitchen and hallway, the same golden window light and color register, the Mother's voice timbre and the hard-cut fragment rhythm. LOCATION MAP the narrow kitchen from @Video1: small table with two chairs against the wall, the window with thin curtains above the sink burning with low evening sun, the hallway of three doors opening off the kitchen, Marcus's bedroom door at its end. CAST Marcus at 16 (@Image1), unrepeatable marker: he receives comfort with his chin down and his eyes up, the same way he argues. the Mother, present only as hands, forearms, a housedress hem, bare feet, a shoulder and hair edge, and a voice that tonight does not crack because she will not allow it. she is never framed whole and her face never resolves. STAGING the wide lens lives inside the embrace-distance, half a meter or closer, so hands and faces bend at the frame edges. across the fragments she advances and he does not retreat, the reverse of every fight. counted exception: in beat six neither of them moves at all. WARDROBE Marcus in the washed teal hoodie over a white tee, grey jeans. the Mother's housedress is the pale florals gone grey with washing from @Video1. no changes within any fragment. GAZE AND EYELINES Marcus looks at her hands in beats one and three, at the floor in beat two, and in beat five his eyes finally lift to where her face would be, just above the frame's edge, and stay there. he never looks at the lens. the camera looks where a son looks: at hands that raised him. EVENT AND SHARED DIRECTION a mother installing one sentence in her son deep enough that it will still be there when he is famous. RULES natural color throughout, 35mm film texture matching @Video1 exactly. hard cuts only. the camera never steadies and never rises above chest height. the mother appears only in parts. every light source renders as a soft, warm, contained glow within its own source. the lens stays super wide and close.
02Beatsone per shot, the part that changes
BEAT 1 — her hands arriving. OPEN super wide inches from Marcus's chest, his hands hanging useless at his sides at frame right, her two hands already entering from frame left and closing around one of his. ACTION her thumbs press across his knuckles, an old habit -> his fingers stay slack a moment, then curl -> the golden window light lies across both their hands like a third hand -> her grip tightens once, deliberate. CAMERA handheld micro-wobble, drifting a palm's width down toward the hands, one move. LENS ECU on hands warping wide at the edges, the kitchen soft gold beyond. PERFORMANCE she wants his hands to stop being fists / blocked by a boy who came home wearing the street / plays certainty like a tool she has used before / turns on his fingers curling. DIALOGUE none. SFX <the kitchen clock, the refrigerator hum> EDIT hard cut on the grip. BEAT 2 — his face, not ready. OPEN Marcus's face fills the super wide frame from centimeters, chin down, eyes down, jaw tight, one cheek gold and one in shadow, a shine at the lower lids he is managing. ACTION his jaw works once on nothing -> his breath drags in through his nose, long -> his eyes close and squeeze and open again wetter -> her shadow crosses the gold on his cheek as she steps nearer. CAMERA locked in trembling hands, no move, the shake is pulse. LENS ECU super wide, his forehead and chin bending at the frame edge. PERFORMANCE wants to say it was not like they said / blocked by a throat that has closed for business / plays composed and is not / turns on her shadow arriving. DIALOGUE none. SFX <a floorboard's slow creak as her weight arrives> EDIT hard cut on the shadow. BEAT 3 — her hands take his face. OPEN both her hands entering the super wide frame from below and left, rising to hold Marcus's jaw on both sides, her wrists and forearms and the frame's edge cutting her at the shoulders, her face above the top of frame, never shown. ACTION her palms settle along his jaw and his whole head lowers a centimeter into them -> her thumb wipes the cheekbone once, taking what was there -> his eyes stay down and she waits -> her hands lift his face one degree, insisting. CAMERA handheld micro-wobble, holding, no move. LENS MCU super wide, boy and hands, her body an out-of-focus warmth at frame left. PERFORMANCE she wants his eyes / blocked by shame that is not his but sits on him anyway / plays patience like iron / turns on the thumb wipe. DIALOGUE her voice above the frame, low, level, each word placed. {Marcus. It's not your fault.} SFX <the pot's low simmer from the stove> EDIT hard cut on the last word. BEAT 4 — the hem, closer than ever. OPEN low at knee height, super wide: her housedress hem and bare feet at frame center, closer to his sneakers than the two of them have stood since the fight, the hallway's long gold shadows behind. ACTION her feet take one final small step until the hem brushes his shin -> his sneakers hold their ground, toes flexing once inside the canvas -> the hem sways and stills -> both their shadows on the floorboards merge into one shape. CAMERA handheld at knee height, sinking one breath lower, one move. LENS WS from the floor, bodies as fragments, the corridor bending away super wide. PERFORMANCE she wants zero distance / blocked by nothing tonight, and that is the event / plays proximity as the whole argument / turns on the shadows merging. DIALOGUE none. SFX <fabric brushing denim, the clock further away now> EDIT hard cut on the merged shadow. BEAT 5 — you hear me. OPEN Marcus's face again, super wide and inches away, but lifted now, eyes finally up to where her face lives just above the frame's top edge, the gold full on him, wet tracks shining and permitted. ACTION his eyes hold on her unseen face for the first time in the memory -> her hand slides from his jaw to the back of his neck -> his chin fights her sentence and loses -> he nods once, small, real, the nod of a contract. CAMERA locked, trembling only with pulse, no move. LENS ECU super wide, his lifted face bending gently at the edges. PERFORMANCE she wants the sentence signed for / blocked by a boy who believes blame keeps him safe / plays the question that is not a question / turns on the nod. DIALOGUE her voice above the frame, harder now, love with its sleeves rolled. {You hear me!} SFX <her hand settling on his neck, cloth-soft> EDIT hard cut on the nod. BEAT 6 — held. OPEN the kitchen from the hallway floor, super wide and low and finally still: the two of them at the window as one shape, her arms around him, his face buried at her shoulder, her head bent to his, both faces unresolved against the burning gold curtains. ACTION steam rises off the pot nobody is watching -> dust turns in the window shaft around them -> his hand comes up her back and grips the grey floral cotton -> neither of them moves again. CAMERA locked, resting on the floor, breath-still. LENS WS deep frame, one shape made of two people, the light the only witness with a whole view. PERFORMANCE the room wants this to be the memory that wins / blocked by all the other memories / plays the silence after the best sentence / turns on his hand gripping her dress. DIALOGUE none. SFX <the pot's simmer, the clock, the street faint through glass> EDIT hold a beat, end.
03Prompt segmentsthe saved look, sound and style, reused across prompts
PHYSICS her thumb on wet skin drags and clears at true friction. the hem sways at cotton weight and settles. his nod carries the small mass of a real head, no more. steam and window dust obey slow indoor air. two bodies leaning into each other carry real weight and settle rather than pose. LIGHTING the same single low evening sun through thin kitchen curtains as @Video1, gold, raking, blooming softly where it strikes, handing the hallway to warm shadow. every light source renders as a soft, warm, contained glow within its own source; faces carry natural skin texture, pores, oil and tear-shine, no gloss. AUDIO <small-apartment room tone, clock, refrigerator hum> <a pot simmering> <street sounds faint through glass> SCORE none STYLE 35mm color film capture, Kodak palette, visible grain, halation on the window highlights, super wide angle lens inches from subjects so the room bends at the edges, handheld throughout except the final locked fragment, natural sources only, memory-real not staged. palette: warm gold 40, deep amber shadow 35, washed teal accents 25. TYPOGRAPHY none VOICE the Mother: American, low and level, the register of a woman who prays and works doubles, tonight refusing to crack because the sentence needs a steady floor; the exclamation lands as pressure, not volume. Marcus at 16: says nothing; his breath and one nod are his whole part. POSITIVE LOCKS the image stays natural 35mm color in every frame and matches @Video1's register exactly. every fragment boundary is a hard splice. Marcus's face matches @Image1 exactly wherever it resolves. the Mother appears only as hands, forearms, hem, bare feet, shoulder and hair edge, and her face never resolves in any frame; her dialogue arrives from above or beside the frame. the lens stays super wide and within arm's reach. the camera never rises above chest height for the full ten seconds.
The Near-Misstake 121.1
An FPV camera flies against downtown traffic and clears the car by inches: one four-second move, rendered silent for sound design. In the film at 1:33 and 1:36.
Show prompt
SCENE CONTEXT A specialist live-action FPV camera travels inches above a rain-darkened downtown Los Angeles avenue against the direction of moving traffic. This is a crowded working city, not an empty automotive stage. Commuter sedans, taxis, a municipal bus, motorcycles, delivery vans, parked cars, pedestrians at a crosswalk, steam from a road grate, patched asphalt, utility covers, stained concrete, mismatched storefronts, old awnings, dirty windows, and practical traffic lights create continuous lived-in depth. The canonical matte blue-grey Lamborghini Temerario from @Image1 begins far down the avenue, partially revealed between ordinary vehicles rather than isolated in the center. The camera maintains one unwavering counterflow trajectory. It passes a taxi, slips beside the bus, collapses distance toward the Lamborghini, clears its driver-side front corner by inches, and continues decisively past it through more traffic. The encounter occupies only a fraction of the route. The camera never brakes, hovers, faces the vehicle for admiration, matches its speed, or ends beside it. Once the Lamborghini flashes past, its rear lights recede behind while the camera continues toward an oncoming delivery van and beneath the changing light of the next intersection. Foreground road texture streaks naturally, traffic produces layered occlusions, and practical lights cause irregular exposure shifts. The Lamborghini remains a heavy road-going object within traffic, while the camera supplies the extreme relative velocity. FORMAT MODE Four-second ultra-wide horizontal live-action single take at twenty-four-frame cadence, one continuous counterflow camera move, one near-miss encounter, and zero cuts. ACTIVE REFERENCES @Image1 -> canonical Temerario and approved downtown Los Angeles world: exact body geometry, matte blue-grey finish, wheels, light signatures, registration, wet road, architecture, and practical night illumination. LOCATION MAP One long avenue holds three active lanes each way. The camera occupies a clear counterflow channel, the Lamborghini approaches in its proper lane, and varied traffic fills near and distant depth. CAST The canonical Temerario is the resolved vehicle. A bus, taxis, commuter cars, motorcycles, and delivery vehicles remain distinct stable traffic participants, while pedestrians stay naturally small at sidewalks and crossings. STAGING The Lamborghini moves steadily with surrounding traffic. The camera begins far away, passes it once at extreme proximity, and continues along the opposing axis without altering velocity. RULES Continuous camera momentum defines the shot. The street remains populated, imperfect, and photographically coherent. All vehicles preserve lane direction, spacing, mass, and stable geometry. BEAT 1 — COUNTERFLOW THROUGH A LIVING CITY OPEN The lens already skims wet patched asphalt as several oncoming vehicles occupy the avenue; the distant Temerario appears between a taxi and bus under irregular practical lighting. ACTION Lane seams fire beneath the lens -> nearby traffic creates close parallax -> the Lamborghini expands through gaps -> the camera clears its front corner -> the car vanishes behind -> forward travel continues through the intersection. CAMERA One continuous ground-skimming counterflow flight with unwavering momentum from opening frame through near-miss and beyond, preserving direction and velocity after the Lamborghini leaves frame. LENS Extreme ultra-wide FPV perspective, one road axis, natural barrel character, heavy foreground displacement, realistic motion blur, and layered depth from moving vehicles at different distances. SFX <silent visual capture prepared for later sound design> EDIT End during continued forward flight after the Lamborghini has fully disappeared behind, with new traffic and the next intersection still rushing toward camera. PHYSICS Vehicles maintain road-going mass, accurate wheel rotation, planted tyre contact, suspension response, restrained wet spray, aerodynamic wake, and credible separation. Camera vibration follows road proximity and turbulent air. LIGHTING Headlights, brake lights, bus interiors, sodium lamps, fluorescent shops, signals, and wet reflections produce localized pools, temporary occlusion, realistic bloom, and imperfect exposure adaptation. AUDIO The generated file is silent for later sound design, while the visible traffic, spray, steam, and pedestrian movement remain synchronized through physical action alone. SCORE none STYLE Photoreal location-shot FPV footage with documentary downtown texture, practical lighting, weathered architecture, imperfect road surfaces, restrained contrast, and palette proportions of wet charcoal 50, matte blue-grey 30, practical amber 20. TYPOGRAPHY none POSITIVE LOCKS The canonical car stays exact and geometrically rigid. Downtown remains busy and visibly inhabited. Camera velocity continues unchanged before, beside, and long after the near-miss.
01Contextwritten once, true for the whole scene
SCENE CONTEXT A specialist live-action FPV camera travels inches above a rain-darkened downtown Los Angeles avenue against the direction of moving traffic. This is a crowded working city, not an empty automotive stage. Commuter sedans, taxis, a municipal bus, motorcycles, delivery vans, parked cars, pedestrians at a crosswalk, steam from a road grate, patched asphalt, utility covers, stained concrete, mismatched storefronts, old awnings, dirty windows, and practical traffic lights create continuous lived-in depth. The canonical matte blue-grey Lamborghini Temerario from @Image1 begins far down the avenue, partially revealed between ordinary vehicles rather than isolated in the center. The camera maintains one unwavering counterflow trajectory. It passes a taxi, slips beside the bus, collapses distance toward the Lamborghini, clears its driver-side front corner by inches, and continues decisively past it through more traffic. The encounter occupies only a fraction of the route. The camera never brakes, hovers, faces the vehicle for admiration, matches its speed, or ends beside it. Once the Lamborghini flashes past, its rear lights recede behind while the camera continues toward an oncoming delivery van and beneath the changing light of the next intersection. Foreground road texture streaks naturally, traffic produces layered occlusions, and practical lights cause irregular exposure shifts. The Lamborghini remains a heavy road-going object within traffic, while the camera supplies the extreme relative velocity. FORMAT MODE Four-second ultra-wide horizontal live-action single take at twenty-four-frame cadence, one continuous counterflow camera move, one near-miss encounter, and zero cuts. ACTIVE REFERENCES @Image1 -> canonical Temerario and approved downtown Los Angeles world: exact body geometry, matte blue-grey finish, wheels, light signatures, registration, wet road, architecture, and practical night illumination. LOCATION MAP One long avenue holds three active lanes each way. The camera occupies a clear counterflow channel, the Lamborghini approaches in its proper lane, and varied traffic fills near and distant depth. CAST The canonical Temerario is the resolved vehicle. A bus, taxis, commuter cars, motorcycles, and delivery vehicles remain distinct stable traffic participants, while pedestrians stay naturally small at sidewalks and crossings. STAGING The Lamborghini moves steadily with surrounding traffic. The camera begins far away, passes it once at extreme proximity, and continues along the opposing axis without altering velocity. RULES Continuous camera momentum defines the shot. The street remains populated, imperfect, and photographically coherent. All vehicles preserve lane direction, spacing, mass, and stable geometry.
02Beatsone per shot, the part that changes
BEAT 1 — COUNTERFLOW THROUGH A LIVING CITY OPEN The lens already skims wet patched asphalt as several oncoming vehicles occupy the avenue; the distant Temerario appears between a taxi and bus under irregular practical lighting. ACTION Lane seams fire beneath the lens -> nearby traffic creates close parallax -> the Lamborghini expands through gaps -> the camera clears its front corner -> the car vanishes behind -> forward travel continues through the intersection. CAMERA One continuous ground-skimming counterflow flight with unwavering momentum from opening frame through near-miss and beyond, preserving direction and velocity after the Lamborghini leaves frame. LENS Extreme ultra-wide FPV perspective, one road axis, natural barrel character, heavy foreground displacement, realistic motion blur, and layered depth from moving vehicles at different distances. SFX <silent visual capture prepared for later sound design> EDIT End during continued forward flight after the Lamborghini has fully disappeared behind, with new traffic and the next intersection still rushing toward camera.
03Prompt segmentsthe saved look, sound and style, reused across prompts
PHYSICS Vehicles maintain road-going mass, accurate wheel rotation, planted tyre contact, suspension response, restrained wet spray, aerodynamic wake, and credible separation. Camera vibration follows road proximity and turbulent air. LIGHTING Headlights, brake lights, bus interiors, sodium lamps, fluorescent shops, signals, and wet reflections produce localized pools, temporary occlusion, realistic bloom, and imperfect exposure adaptation. AUDIO The generated file is silent for later sound design, while the visible traffic, spray, steam, and pedestrian movement remain synchronized through physical action alone. SCORE none STYLE Photoreal location-shot FPV footage with documentary downtown texture, practical lighting, weathered architecture, imperfect road surfaces, restrained contrast, and palette proportions of wet charcoal 50, matte blue-grey 30, practical amber 20. TYPOGRAPHY none POSITIVE LOCKS The canonical car stays exact and geometrically rigid. Downtown remains busy and visibly inhabited. Camera velocity continues unchanged before, beside, and long after the near-miss.
Characters
Every face in the film, each from one sheet. Copy the prompt that made a sheet; for Marcus and Ray, copy the cast token that goes into every video prompt they are in.
Marcus34.1
The final sheet: locs, diamond grill, pearls over the iced chain, the patchwork tracksuit. It rides every Marcus prompt as @Image1.
Image 1: exact character identity to preserve. Images 2, 3, and 4: grill and jewelry reference only. Image 5: patchwork outfit logic reference only. Create a custom three-panel character sheet on a neutral warm-gray seamless background with no text, no labels, and no rulers. The left 55% of the frame is one dominant head-and-shoulders portrait, and the right 45% is split into exactly two full-body views (front three-quarter and back three-quarter). The subject is Marcus, a late-20s Black man with a tall wiry build, long angular face, high cheekbones, short freshly retwisted twists, a pale scar through the outer third of his left eyebrow, and visibly faded old tattoos on his hands and neck. In the portrait, he gives a restrained half-smile with lips parted to reveal a full upper-and-lower diamond and gold grill. He is wearing a deep petrol-teal technical track jacket and relaxed trousers with asymmetric paisley bandana patchwork panels in turquoise, midnight navy, coral-red, and cream, along with spotless white leather low-tops. He wears layered chains outside the jacket including a diamond chain and a pearl strand, diamond stud earrings, thick iced rings, and a small nose ring. Shot in hyperreal casting photography style with soft frontal key lighting, narrow edge light, honest skin pores, and no glossy AI retouching.
Image 1: exact character identity to preserve. Image 2, 3, 4: grill and jewelry reference. Image 5: outfit reference. Create a premium custom THREE-PANEL character sheet on a 16:9 landscape canvas where the left 55% is one dominant head-and-shoulders portrait and the right 45% is split into exactly two full-body views (front three-quarter and back three-quarter, head to shoes), with no additional portraits, no inset details, no labels, no text, and no rulers. The subject is Marcus, a late-20s Black man with a tall wiry narrow-shouldered build, long angular face, high cheekbones, narrow tapered jaw, deep-set still eyes, short freshly retwisted twists with a razor lineup, a pale scar through the outer third of his left eyebrow, and old visibly faded tattoos on his hands and neck. He is wearing bold modern rapper sportswear consisting of a saturated lacquer-red technical track jacket and relaxed matching trousers with asymmetric paisley and bandana patchwork panels in burgundy, navy, cream, and small forest-green accents, made of premium matte nylon, paired with spotless white leather low-tops. He wears layered chains outside the jacket including one diamond chain and one fine pearl strand, diamond stud earrings, thick iced rings on both hands, an understated bracelet, and a small nose ring. In the large portrait, he gives a restrained half-smile with lips parted just enough to unmistakably show a full upper-and-lower grill with a warm gold framework and tightly set white diamonds, without using a finger to pull the lip. The visual style is hyperreal casting photography against a neutral warm-gray seamless studio background, lit with a soft frontal key and narrow edge light, featuring honest skin pores and texture, no glossy AI skin, no beauty retouching, an 85mm portrait feel on the face and 50mm full-body feel, restrained editorial color, and clean graphic spacing between the three panels with no overlap, ensuring no celebrity likenesses from the references are copied.
MARCUS, a Black man in his late twenties, lean, mid-brown skin with visible pore texture.
Shoulder-length dark locs hanging forward on both sides of his face, the temples faded short.
A full DIAMOND GRILL across his upper teeth. A small silver ring in his left nostril and a
diamond stud in each ear. Fine-line tattoos across both sides of his neck and his upper chest.
A single strand of white pearls worn over a heavy iced Cuban-link chain. Thin moustache and a
sparse chin beard.
Image 1: adult identity anchor. Image 2: base image to modify. Change: Convert the entire image from black-and-white to natural restrained color with truthful dark-brown skin, an off-white cotton T-shirt, faded charcoal jeans, and neutral gray sneakers, and completely remove the eyebrow scar so both eyebrows are fully intact, solid, and natural with no cuts or gaps. Preserve: The exact three-panel 55/45 layout, the child's facial features, hairstyle, quiet expression, poses, lighting direction, and neutral studio background from Image 2. Constraints: Absolutely no scar or notch in the eyebrows, no monochrome, no tattoos, no jewelry, no logos, no text.
Image 1: exact character identity to age-regress to 8 years old. Create a custom three-panel character sheet on a neutral plain studio background with no text, no labels, and no rulers, where the left 55% of the frame is one dominant head-and-shoulders portrait and the right 45% is split into exactly two full-body views (front three-quarter and back three-quarter) of Marcus at exactly eight years old. The subject is a young Black boy with a slim child's build, rounder cheeks, shorter jaw, larger child-proportioned eyes, short natural twists, a pale scar through the outer third of his left eyebrow, and a quiet guarded expression, wearing a simple cheap white T-shirt, slightly worn dark jeans, and scuffed everyday sneakers. Shot in black-and-white 16mm casting photography style with lifted blacks, visible heavy grain, soft window-like frontal light, modest contrast, honest skin texture, and slight 28mm intimacy, ensuring no tattoos, no jewelry, no piercings, no grill, no logos, no adult styling, and no text anywhere in the frame.
Image 1: base image to preserve. Everything in this image must remain identical except for the boy's left eyebrow, which must be fully intact, naturally continuous, and unmarked with no scar, cut, notch, or gap in all three panels. Preserve the exact age-eight identity, face, short twists, expression, child anatomy, pose, off-white T-shirt, charcoal jeans, gray sneakers, natural color, skin tone, lighting, grain, neutral studio background, framing, three-panel layout, and every other element exactly as they appear in Image 1.
Marcus at sixteen36.3
The washed teal hoodie he wears in the kitchen flashback.
Image 1 is the base image to preserve and edit; everything in this image must remain identical except for the wardrobe, which should be changed in all three panels to a washed muted teal zip hoodie worn open over a plain off-white T-shirt, loose faded black jeans, and worn charcoal sneakers that are simple, inexpensive, and logo-free, while strictly preserving the exact three-panel layout, neutral studio background, lighting, and the boy's exact identity, 13-14 year old adolescent anatomy, short twists, quiet expression, and the pale scar through the outer third of his left eyebrow.
Image 1: exact character identity to preserve and age-regress. Create a monochrome three-panel casting reference for Marcus at thirteen to fourteen years old on a neutral plain studio wall and floor, with no text, no labels, and no rulers. The left 55% of the frame is one dominant head-and-shoulders portrait, and the right 45% is split into exactly two full-body views (front three-quarter and back three-quarter). The subject is a 13-14 year old Black boy, age-regressed from Image 1, preserving his deep-set still eyes, emerging high cheekbones, developing long angular face, short natural twists, and the pale scar through the outer third of his left eyebrow. He has a slim adolescent frame, narrow shoulders, softer jaw and cheeks, and absolutely no beard or mustache. He wears a cheap white T-shirt, loose faded dark jeans, and worn sneakers, with no tattoos, no jewelry, no piercings, and no grill. His expression is quiet and watchful. Shot in black-and-white 16mm casting photography style with lifted blacks, visible heavy grain, soft window-like frontal light, modest contrast, and honest adolescent skin texture.
Image 1: exact character identity to preserve and age-regress. Image 2: exact layout, poses, adolescent anatomy, hairstyle, expression, and left-eyebrow scar to preserve. Create a custom three-panel character sheet in natural restrained color on a neutral plain studio wall and floor, with no text, no labels, and no rulers. The left 55% of the frame is one dominant head-and-shoulders portrait, and the right 45% is split into exactly two full-body views (front three-quarter and back three-quarter). The subject is Marcus, a 13-14 year old Black boy, age-regressed from Image 1, preserving his deep-set still eyes, emerging high cheekbones, developing long angular face, short natural twists, and the pale scar through the outer third of his left eyebrow. He has a slim adolescent frame, narrow shoulders, softer jaw and cheeks, and absolutely no beard or mustache. He wears a plain off-white cotton T-shirt, loose faded charcoal jeans, and worn neutral gray sneakers, with no tattoos, no jewelry, no piercings, and no grill. His expression is quiet and watchful. Shot in natural restrained color with truthful dark-brown skin tones, neutral studio tones, soft window-like frontal light, modest contrast, subtle photographic grain, and honest adolescent skin texture.
Ray38.1
The final sheet: a beard greying at the chin, the navy work polo. It rides every Ray prompt as @Image2.
Image 1: reference for shared genetic geometry only. Create a custom three-panel character sheet on a neutral warm-gray seamless background with no text, no labels, and no rulers, where the left 55% of the frame is one dominant head-and-shoulders portrait and the right 45% is split into exactly two full-body views (front three-quarter and back three-quarter) of Ray, a 32-year-old Black man who shares the deep-set eyes, straight prominent nose, high cheekbones, and long facial geometry of the man in Image 1 but is a distinct older brother with a slightly broader jaw, softer unshaped natural hairline, short textured hair, calm alert eyes, and a close predominantly dark beard with a tiny trace of grey at the chin. Ray has a lean, sturdy working-man build and wears a washed navy logo-free work polo, charcoal heavy work trousers, cared-for brown work boots, a worn gold wedding band, and a rubber-strap digital watch, with a settled and unsentimental expression, completely lacking the eyebrow scar, tattoos, grill, nose ring, and jewelry of Image 1. Shot in natural restrained-color casting photography with soft frontal light, honest skin pores, fabric wear, and no glossy retouching, maintaining one consistent identity, body, beard, wardrobe, and accessories across all three panels.
Image 1: mandatory character-sheet layout to preserve. Image 2: Marcus, use only to inherit family facial structure for Ray. Make a character sheet as per Image 1 for Ray, a Black man in his mid thirties with a thickset, broad build softened through the middle, preserving the exact nose family, eye placement, brow spacing, and underlying craniofacial structure from Image 2 but aged and differentiated naturally, featuring a soft unshaped hairline without a lineup, a full beard with grey only at the chin, and a calm, settled face, wearing a navy work polo softened by washing, heavy work trousers with real creasing behind the knees, cared-for brown work boots, a worn gold wedding band, and a rubber-strap digital watch. Visual Style for the character: Flat neutral production lighting, natural pores and fabric wear, documentary-grade hyperreal photography. Fill the info block exactly as: NAME: "RAY" AGE: "ADULT" HEIGHT: "72\" (inches)" WEIGHT: "" GENDER: "MALE" Replace each ruler's `X" (inches)` placeholder with "72\" (inches)", and set the figure against the ruler to match — don't default to the top. No extra captions, scenery, logos, watermark, or duplicate figures.
Image 1: mandatory character-sheet layout. Image 2: Marcus, used only for inherited family geometry (nose, eye placement, brow spacing). Make a character sheet as per Image 1 for Ray, a Black man in his mid thirties, 73 inches tall, broad and thickset with a comfortably softened middle, featuring a soft natural hairline without sharp grooming, and a full beard with grey only through the chin, wearing a navy work polo softened by many washes, heavy work trousers with real knee creases, worn but cared-for brown work boots, a worn gold wedding band, and a rubber-strap digital watch. Visual Style for the character: hyperreal photographic fidelity, flat neutral production lighting, honest skin and material texture. Fill the info block exactly as: NAME: "RAY", AGE: "ADULT", HEIGHT: "73 INCHES", GENDER: "MALE". Replace each ruler's X" (inches) placeholder with "73\" (inches)", and set the figure against the ruler to match — don't default to the top.
Image 1: base layout to preserve. Image 2: Marcus, the younger brother, for facial inheritance only. Make a character sheet as per Image 1 for Ray, a Black man in his mid thirties, thickset, broad-chested and softened around the waist, sharing the same eye placement, brow spacing, nasal bridge family, and underlying facial proportions as the man in Image 2 but clearly distinct, older, and heavier, with a natural unshaped hairline and a full beard with grey confined to the chin, wearing a softened navy work polo, heavy work trousers with authentic creasing, cared-for brown work boots, a worn gold wedding band, and a rubber-strap digital watch. Visual style is a hyperreal production-reference image with flat neutral production light, real pores, beard texture, and garment wear. Fill the info block exactly as: NAME: "RAY" AGE: "ADULT" HEIGHT: "72\" (inches)" GENDER: "MALE" Replace each ruler's 'X\" (inches)' placeholder with "72\" (inches)", setting the figure against the ruler to match without defaulting to the top.
Image 1: layout reference to preserve exactly. Image 2: facial structure reference (younger brother). Make a character sheet as per Image 1 for Ray, a dark-skinned Black man in his mid thirties with a thickset and broad build with a softened middle, sharing the exact nose family, wide eye placement, brow spacing, and foundational face geometry of the man in Image 2 but as a distinct older sibling, featuring a soft natural hairline without lineup, a full beard with grey only at the chin, and a warm but unsentimental, slow, settled expression, wearing a navy work polo softened from washing, heavy work trousers with real creasing, worn and carefully maintained brown work boots, a worn gold wedding band, and a rubber-strap digital watch, shot in flat neutral production light with honest skin, beard and fabric texture, and hyperreal photography. Fill the info block exactly as: "NAME: RAY", "AGE: ADULT", "HEIGHT: 71 INCHES", "GENDER: MALE", and replace each ruler's placeholder with "71 INCHES", setting the figure against the ruler to match.
RAY, a Black man in his mid thirties, thickset and settled, darker skin than his brother. The
SAME nose and the SAME eyes — they read as blood on sight. Close-shaved head, a beard going grey
at the chin ONLY, a pale scar splitting the right eyebrow. A plain wedding ring, and no other
jewellery. Clean work clothes at the end of a shift — worn, well kept, and NEVER shabby.
Make a hyperreal character sheet exactly according to Image 1 for Ray at nineteen years old, using Image 2 as the sole identity and aging reference: preserve Ray’s recognizable long facial geometry, eye placement, brow spacing, nose family, cheekbones, mouth, deep brown complexion, and grounded older-brother presence, naturally de-aged into a believable late teenager whose face has not yet fully hardened into adulthood. Keep his exact identity distinct from Marcus. Give him short natural hair, faint adolescent facial-hair shadow only, unretouched skin, and a lean sturdy build with restrained physical confidence. Dress him consistently in every panel in a washed navy work shirt worn open over a plain white ribbed undershirt, loose dark denim, a simple leather belt, and worn work boots, all modest early-2000s clothing with plain surfaces and no branding. His posture should suggest a nineteen-year-old already carrying responsibility: shoulders level, jaw quiet, hands relaxed, watchful rather than posed. Render one dominant face close-up plus FRONT, BACK, LEFT PROFILE, and RIGHT PROFILE full-body views, anatomically consistent and evenly lit against the template’s neutral studio background. Hyperreal casting photography, truthful Black skin, visible pores and minor adolescent texture, restrained natural color, soft frontal illumination, and zero glamour treatment. Fill the information block exactly as: NAME: RAY; AGE: LATE TEEN; HEIGHT: 71" (inches); WEIGHT: 162 lb; GENDER: MAN. Replace every ruler placeholder with 71" (inches), keeping figure scale consistent. No additional text and no duplicate text.
Make a hyperreal character sheet exactly according to Image 1 for Ray at eleven years old, using Image 2 as the sole identity and aging reference: preserve Ray’s recognizable long facial geometry, eye placement, brow spacing, nose family, cheekbones, mouth, deep brown complexion, and distinct older-brother presence, naturally de-aged into a believable preteen rather than miniaturizing an adult. His eyebrows are intact, his face has no facial hair, and his short natural hair is period-authentic for a modest late-1990s household. Dress him consistently in every panel in a faded navy athletic T-shirt, loose worn denim shorts, white crew socks, and inexpensive scuffed sneakers, with plain surfaces and no branding. He has a lean, sturdy preteen build, slightly taller and broader than eight-year-old Marcus would be, with natural child anatomy and an alert protective stillness already visible in his posture. Render one dominant face close-up plus FRONT, BACK, LEFT PROFILE, and RIGHT PROFILE full-body views, anatomically consistent and evenly lit against the template’s neutral studio background. Hyperreal casting photography, truthful Black skin, visible pores, modest wardrobe wear, restrained natural color, soft frontal illumination, and zero glamour treatment. Fill the information block exactly as: NAME: RAY; AGE: PRETEEN; HEIGHT: 57" (inches); WEIGHT: 78 lb; GENDER: BOY. Replace every ruler placeholder with 57" (inches), keeping figure scale consistent. No additional text and no duplicate text.
The mother119.1
Their mother: the sage housedress from the flashbacks.
Make a hyperreal character sheet using Image 1’s exact five-panel layout for Marcus’s mother, whose visible identity is designed through credible maternal family resemblance to the adult Marcus in Image 2: preserve his deep brown complexion, long angular facial family, high cheekbones, deep-set almond eyes, distinctive brow spacing, slender nose bridge, full defined lips, and narrow jaw lineage, transformed naturally into a unique young Black mother rather than a female duplicate. Her face is fully visible on this private continuity sheet only. Give her short natural coily twists, an unretouched face with pores and subtle fatigue, a slender practical build, and a protective, resilient stillness. Dress her consistently in every view in a modest, faded sage-green patterned cotton house dress inspired by Image 3, with inexpensive construction, softened fabric, and period-authentic late-1990s domestic wear; Image 4 influences only the grounded maternal presence and short natural hair, never its exact identity. Render one face close-up plus FRONT, BACK, LEFT PROFILE, and RIGHT PROFILE full-body views, all anatomically consistent, evenly lit against a neutral warm-gray studio background. Hyperreal casting photography, truthful Black skin, restrained color, soft frontal key, subtle edge separation, natural fabric wear, zero glamour retouching. Fill the information block exactly as: NAME: MOTHER; AGE: YOUNG ADULT; HEIGHT: 66" (inches); WEIGHT: 125 lb; GENDER: WOMAN. Replace every ruler placeholder with 66" (inches), keeping figure scale consistent. No additional text, no duplicate text.
Early explorations21.1
Marcus before the look was found: all-black model sheets.
Make a character sheet as per Image 1 for Marcus, a Black man in his late twenties with a tall, wiry build, narrow shoulders, a long severe face, high cheekbones, narrow tapered jaw, slightly asymmetric broad nose, deep-set motionless eyes, prominent ears, and short freshly retwisted twists with a razor lineup, featuring a pale scar through the outer third of his left eyebrow and faded old tattoos on his hands and neck, wearing a matte-black logo-free technical jacket with no creasing, a pristine black tee, a chain hidden beneath the shirt with only rear neck links visible, a small nose ring, a removable gold grill, and immaculate white leather low-tops. Visual Style for the character: Flat neutral production lighting, natural pores, minor facial asymmetry, documentary-grade hyperreal photography, conveying contained damage and exhausted vigilance with real gangster-rapper credibility without costume or stereotype. Fill the info block exactly as: NAME: "MARCUS" AGE: "YOUNG ADULT" HEIGHT: "74\" (inches)". Fills the info-block field and the ruler labels — keep them consistent. WEIGHT: "" GENDER: "MALE" Replace each ruler's `X" (inches)` placeholder with "74\" (inches)", and set the figure against the ruler to match — don't default to the top. No additional captions, logos, watermark, scenery, props, or duplicate figures.
Make a character sheet as per Image 1 for Marcus, a Black man in his late twenties with a lean, narrow, wiry build, featuring a long elegant face, unusually high cheekbones, hooded deep-set almond eyes, a slightly crooked narrow nose, subtle under-eye hollows, a tapered jaw, short freshly retwisted twists with a razor lineup, a pale scar through the outer third of his left eyebrow, and faded old hand and neck tattoos, wearing matte-black logo-free technical outerwear, a pristine black tee with a gold chain concealed beneath the shirt showing only links at the rear neck, a small nose ring, a removable gold grill, and unscuffed white leather low-tops. Visual Style for the character: hyperreal photographic fidelity, flat neutral production lighting, natural pores and asymmetry, understated rapper-gangster authority, never fashion-model polish or stereotype. Fill the info block exactly as: NAME: "MARCUS", AGE: "YOUNG ADULT", HEIGHT: "74\" (inches)", WEIGHT: "165 lbs", GENDER: "MALE". Replace each ruler's X" (inches) placeholder with "74\" (inches)", and set the figure against the ruler to match — don't default to the top.
Image 1: base layout to preserve. Make a character sheet as per Image 1 for Marcus, a Black man in his late twenties, wiry and narrow through the shoulders, with an elongated angular face, broad nasal bridge with an old healed break, close-set deep eyes, high temples, sharp cheek planes, narrow jaw, short freshly retwisted twists with a precise lineup, a pale scar through the outer third of his left eyebrow, and old faded tattoos on his hands and neck, wearing a matte-black logo-free technical jacket, new black tee, chain hidden under the shirt with only rear neck links visible, small nose ring, removable gold grill, and immaculate white leather low-tops. Visual style is a hyperreal production-reference image with neutral production lighting, honest pores, healed skin texture, and quiet street credibility. Fill the info block exactly as: NAME: "MARCUS" AGE: "YOUNG ADULT" HEIGHT: "75\" (inches)" GENDER: "MALE" Replace each ruler's 'X\" (inches)' placeholder with "75\" (inches)", setting the figure against the ruler to match without defaulting to the top.
Image 1: layout reference to preserve exactly. Make a character sheet as per Image 1 for Marcus, a dark-skinned Black man in his late twenties with a lean and deceptively slight build, a long oval face, pronounced brow ridge, widely spaced deep-set eyes, flattened upper ears, a strong narrow nose, fine mouth, tapered jaw, short freshly retwisted twists with a razor lineup, a pale scar through the outer third of his left eyebrow, and old faded hand and neck tattoos, wearing matte-black logo-free technical outerwear, a pristine black tee, a chain concealed under the shirt with only links visible at the back of the neck, a small nose ring, a removable gold grill, and immaculate white leather low-tops, shot in flat neutral production light with natural pores, facial asymmetry, and hyperreal photography conveying stillness and spiritual fatigue. Fill the info block exactly as: "NAME: MARCUS", "AGE: YOUNG ADULT", "HEIGHT: 73 INCHES", "GENDER: MALE", and replace each ruler's placeholder with "73 INCHES", setting the figure against the ruler to match.
The car
One canonical sheet, attached to every render the car is in. Seedance reads the geometry and the plate off the image; nobody asks it for them in words.
The Lamborghini66.5
The canonical car: four views at night, with the California plate. Attached to every car render.
Everything in Image 1 must remain identical except the vehicle license-plate surfaces. Replace the plain dark plate surfaces with physically mounted California registration plates, using the same believable invented registration in every visible view. The plates follow real California proportions and restrained white styling, sit correctly inside the existing front and rear mounting recesses, obey each panel's perspective, exposure, reflections, occlusion, and depth of field, and look naturally photographed as part of the car. Preserve every other pixel-level visual decision: the exact four-panel layout and dividers, exact vehicle geometry, pale metallic blue-silver paint, wheels, ride height, active lighting signatures, dark tinted glazing, two concealed occupant silhouettes, camera positions, downtown Los Angeles setting, wet asphalt, anamorphic plate character, contained practical glows, grain, color, contrast, architecture, and composition. No redesign, reframing, added objects, extra graphics, captions, badges, or text anywhere outside the license plates.
Create a photoreal four-panel automotive identity sheet showing the single exact vehicle documented by Images 1 and 2. Image 1 has highest authority over the complete front identity: nose proportions, hood geometry, slim upper lamps, lower hexagonal DRLs, windshield shape and front wheel placement. Image 2 has equal authority over the side and rear identity: roofline, wheelbase, door cuts, side intake, rear hips, deck, fascia, wheels and taillights. Reconcile both references into one mechanically identical vehicle across every panel, preserving their exact production geometry rather than interpreting the design. Arrange four equal panels with thin neutral dividers: low front three-quarter, level full side profile, low rear three-quarter, and high front three-quarter. The car has the same pale metallic blue-silver finish, identical wheels, ride height, glasshouse, body cuts and lighting hardware in every panel. All slim upper headlights and both lower hexagonal DRLs are complete, active and balanced whenever visible; rear lamps are active whenever visible. Exterior glazing is realistic dark factory tint: side and rear glass appear nearly opaque, while the windshield remains heavily shaded with only two restrained human silhouettes visible inside. The left-hand-drive driver silhouette sits correctly behind the steering wheel and one passenger silhouette occupies the opposite front seat; facial features, hands and wardrobe remain concealed by tint and shadow. Downtown Los Angeles at night, wet-black asphalt, sparse cyan traffic signals and restrained sodium practicals. Every practical light renders as a soft, round, contained glow held within its physical source. Strong anamorphic plate character through horizontal compression, oval depth rendering, curved barrel-edge distortion and subtle edge chromatic aberration. Fine 500T grain, deep blacks with readable body edges, restrained highlight rolloff, honest road grime and photographic exposure. Plain dark license-plate surfaces and plain surrounding architecture keep the vehicle geometry dominant. The four panels function as a precise downstream identity reference rather than advertising imagery.
Create a rigorous 16:9 automotive production-reference sheet for a real Ferrari SF90 Stradale in an understated factory gloss-black specification. Pure light-grey catalog background with thin dividers. Show the identical car in six consistent panels: front orthographic, rear orthographic, left side profile, front three-quarter, rear three-quarter, and a wide cockpit view showing both seats, steering wheel, center console, dashboard and instrument cluster. Preserve authentic production geometry, wheelbase, vents, lighting signatures, wheel design, glasshouse, cabin architecture and manufacturer badging. Neutral three-point studio lighting, 50mm perspective, realistic paint, rubber, glass and carbon texture, no dramatic beauty lighting. Small header only: EXACT TEXT: "FERRARI SF90 STRADALE". No additional text, specifications, people, road scenery, invented body kit, watermark, or duplicate vehicle.
Create a rigorous 16:9 automotive production-reference sheet for a real McLaren 750S coupe in understated factory deep graphite metallic on a pure light-grey catalog background with thin dividers. Show the identical car in six consistent panels: front orthographic, rear orthographic, left side profile, front three-quarter, rear three-quarter, and a wide cockpit view showing both seats, steering wheel, center console, dashboard and instrument cluster. Preserve authentic production geometry, compact proportions, intake forms, lighting signatures, wheel design, glasshouse, dihedral-door cabin architecture and manufacturer badging. Use neutral three-point studio lighting, 50mm perspective, realistic paint, rubber, glass and carbon texture, with no dramatic beauty lighting. Include a small header with EXACT TEXT: "McLAREN 750S". Constraints: No additional text, no specifications, no people, no road scenery, no invented body kit, no watermark, and no duplicate vehicle.
A rigorous 16:9 automotive production-reference sheet for a real Aston Martin Valkyrie coupe in an austere factory exposed dark-carbon specification, set against a pure light-grey catalog background with thin dividers. Show the identical car in six consistent panels: front orthographic, rear orthographic, left side profile, front three-quarter, rear three-quarter, and a wide cockpit view showing both seats, steering wheel, center console, dashboard and instrument cluster. Preserve authentic production geometry, extreme venturi tunnels, narrow cabin, lighting signatures, wheel design, glasshouse, cockpit architecture and manufacturer badging with neutral three-point studio lighting, 50mm perspective, realistic carbon, rubber and glass texture, and no dramatic beauty lighting. Include a small header with EXACT TEXT: "ASTON MARTIN VALKYRIE" and absolutely no additional text, specifications, people, road scenery, invented aero components, watermarks, or duplicate vehicles.
Locations and look
The worlds the film moves through, and the plates that set its light.
Motion-safe cinematic location plate for video handoff. A two-lane road at night runs dead straight away from camera into darkness. The last city glow sits low on the horizon behind low hills. One distant light waits far down the road. Dry asphalt, faded centre line, scrub on both sides, no buildings or lamps. Palette: 70 near-black asphalt and sky, 20 dim horizon glow, 10 cold starlight. Photographed on film with heavy real grain and deep falloff. No vehicles or people. MODERATE anamorphic lens character: horizontal squeeze and compression, oval elliptical bokeh, horizontally stretched highlights, curved barrel-edge distortion, chromatic aberration toward the edges. NO lens flares, NO light streaks, NO floating bokeh orbs. 2.39:1.
The cabin54.2
The empty passenger seat of the cockpit, lit only by the cluster, with room for a driver.
Motion-safe plate from the empty passenger seat of the exact Lamborghini Revuelto cockpit in Image 1. Preserve its steering wheel, cluster, dashboard, centre console, carbon and black leather. Camera sits low beside the passenger headrest, angled toward the driver position and windscreen, leaving room for a driver to be added. Cold cyan cluster underlight is the only cabin illumination; weak sodium streetlight rakes through the windscreen. An empty residential road recedes outside. Deep matte shadow dominates. Film grain, imperfect focus, worn surfaces. Every defocused streetlamp is a wide horizontal ellipse, approximately 2.5 times wider than tall. Highlight blooms remain attached to their physical sources with clean dark space between them. Moderate horizontal compression, curved edge distortion and edge chromatic aberration. Clean glass. Empty cabin. 2.39:1.
The tunnel55.3
An empty tiled tunnel, sodium fixtures along the crown.
Static fast-shutter film photograph from a camera fixed at road centre inside an empty tiled Los Angeles tunnel. Crisp dry lane markings lead to a small dark mouth far ahead. White tile walls and ceiling remain mostly in deep shadow. A sparse row of rectangular sodium fixtures sits along the crown; the nearest housings are sharp, while only the most distant fixtures soften into compact flattened oval blooms, each wider than tall and separated by black gaps. No motion exposure. Near-black dominates, with restrained sodium and a small cold exit point. Fine grain, imperfect edge focus, mild barrel curvature and edge chromatic aberration. Clean dark glass. 2.39:1.
The old street56.2
An empty residential lane at night, three inches above the wet asphalt.
Motion-safe night location plate, camera three inches above wet asphalt on an empty residential travel lane. Single-storey houses and chain-link fencing sit in deep shadow; parked cars remain tight to both kerbs. Sparse sodium lamps recede down the street and reflect as muted horizontal bands in the road. Near-black asphalt occupies most of frame; sodium is limited to isolated sources and reflections, with a few cold window accents. Heavy film grain, imperfect focus. Every defocused lamp is a wide horizontal ellipse, approximately 2.5 times wider than tall. Each highlight bloom stays compact and attached to its lamp, with uninterrupted darkness around it. Strong horizontal compression, curved barrel-edge distortion and edge chromatic aberration. Clean glass. Empty lane, nobody present. 2.39:1.
Downtown57.3
An empty downtown canyon at night: wet asphalt between dark towers.
Static fast-shutter film photograph from road level in an empty downtown Los Angeles street canyon at night. Wet asphalt runs deep between dark glass towers; palms interrupt the skyline; one distant tower crown glows through haze. Near-black architecture and road dominate. Only a few nearby sodium fixtures remain sharp. Distant lamps and windows soften into compact flattened oval blooms, each clearly wider than tall and separated by darkness. Reflections stay dim, local and horizontal beneath their physical sources. Restrained cold office light above, sparse warm pools below. Fine grain, atmospheric haze, imperfect edge focus, mild barrel curvature and edge chromatic aberration. Clean dark glass. Empty road and pavements. 2.39:1.
Driver plate39.4
Marcus at the wheel, the teal city behind the glass: a still made to hand to video.
Motion-safe film still for video handoff. Image 1: exact character identity, skin tone, and wardrobe to preserve. Image 2: composition, exposure, and lighting logic reference only. Subject: Marcus, a late-20s Black man with a long angular face, high cheekbones, short twists, and a pale scar through the outer third of his left eyebrow, wearing a deep petrol-teal track jacket. Action: Sitting in the driver's seat of a dark modern supercar at night, facing forward in close profile with a fixed, unreadable expression. Scene: Inside a dark car cabin at night, cool cyan street bokeh visible beyond the side window, subtle glass reflections. Blocking: Close profile, positioned left of center, tight framing on the face and window to exclude the car roof and sun visor. Lighting: Warm dashboard light passing narrowly over his cheek, nose, and lips. Deep cabin blacks. Style: Photoreal cinematic still, clean modern cinema sensor, natural restrained color, honest skin pores, controlled highlights. Constraints: No text anywhere in the frame. No signage, no logos, no UI labels, no printed matter, no graffiti, no random letters. All surfaces that could carry text should be plain. No baked film stock, no grain, no halation, no VHS damage, no glamour retouching. Do not copy the person or vehicle identity from Image 2.
Master plate39.6
Marcus stripped back to neutral: no jewellery, no grill, the teal jacket, the night outside.
Motion-safe clean neutral master plate. Reference priority: Image 1 is the sole authority for exact character identity, face, pale scar through the outer third of his left eyebrow, short twists, skin tone, and proportions. CRITICAL: Marcus wears ABSOLUTELY NO JEWELRY. Bare neck, bare ears, bare nose, and natural teeth with no grills. Strip all accessories seen in Image 1. He wears a plain deep petrol-teal track jacket with no patterns. Image 2 controls only camera composition, exposure, and lighting logic; do not copy its person. Frame Marcus left of center in a tight side profile, looking straight through the windshield with an unreadable expression. A narrow restrained amber dashboard pass touches only his cheek, nose, and lips. Cool cyan street bokeh remains outside the side window. Preserve deep cabin blacks with usable shadow detail and subtle clean glass reflections. Render as an untouched modern cinema-sensor exposure with natural pores and neutral Black skin.
Dialogue plate61.4
Both brothers in the cabin: the source plate for the dialogue scenes.
Image 1: Marcus identity, build, scar, hair, tattoos, and facial structure. Image 2: Ray identity, age, beard, build, family geometry, and workwear. Image 3: Lamborghini Revuelto cabin materials, dashboard, seats, and center-console geometry. Create a 16:9 source plate for a dialogue scene. Marcus (Image 1) sits at the wheel on frame left, holding the wheel with restrained posture. He wears matte black technical outerwear over a black tee. His gold chain is completely hidden under his shirt, not visible on his chest. He wears a gold grill and a nose ring. Ray (Image 2) sits in the passenger seat on frame right, sitting comfortably with hands low. He wears a washed navy logo-free work polo. Ray wears a single rubber-strap digital watch on his left wrist ONLY; his right wrist is bare. Both men share the same nose and eye placement, reading as brothers, while their age, build, hair, beard, and wardrobe remain distinct. Both men face straight through the windshield, not looking at each other. They are divided by the tall transmission tunnel and center console (Image 3). The camera is mounted on the dashboard, providing a balanced two-shot where both mouths and torsos are readable. The cabin is sealed, and the night beyond the glass is nearly black. The cabin is illuminated EXCLUSIVELY by cold cyan instrument-cluster light rising from below, lighting their faces from the dashboard. The ceiling of the car is completely black and invisible in the shadow. There are absolutely no overhead lights, dome lights, or ceiling lights. There is no outside streetlight. Moderate anamorphic lens character: horizontal squeeze, oval elliptical bokeh, horizontally stretched source-bound highlights, curved barrel-edge distortion, chromatic aberration toward the edges. Natural skin, fine film grain, clean glass, no glossy AI retouching.
Storyboards
Keyframe strips that planned the flashbacks, and pre-vis grids that tested blocking and rhythm before any video was rendered.
Laundry keyframes125.1
Marcus at eight and his mother at the washtub. Her face stays hidden in every panel.
Create a single hyperreal four-panel cinematic keyframe strip for an eight-second childhood memory. Image 1 is the exact identity of eight-year-old Marcus. Image 2 is the private identity and wardrobe reference for his young mother, but in every finished-film panel her face remains concealed through profile crop, foreground objects, hair, gesture, or shallow focus. Image 3 defines the sunlit laundry-world composition, humble domestic texture, natural backlight, and intimate low perspective. Image 4 influences maternal stillness and short natural hair only. Late-1990s poor but happy home where routine work becomes a game: cramped weathered kitchen opening into a small laundry yard, chipped enamel, hand-washed clothes, faded curtains, repaired furniture, everything clean and cared for. Four sequential beats: Marcus stands on an upturned crate helping stir soapy clothes while Mother steadies him at the waist; he flicks a small crown of soap bubbles toward her and she recoils with restrained laughter; they carry a wet sheet together into the yard, the fabric briefly billowing over both of them like a tent; final floor-level fragment beneath the sheet as their feet shuffle in a tiny dance and her hand reaches down for his. Loose intimate 28mm photography, asymmetric framing, natural restrained color, warm window light, truthful dark skin, lifted shadows, tactile wear, slight exposure breathing, unretouched documentary texture. Each panel is motion-safe with clear spatial depth, readable limbs, and room for natural continuation. Preserve both identities, ages, hair, body proportions, and modest wardrobe exactly from their references. Marcus’s eyebrows are intact at age eight. Mother’s full face is never visible. No captions, borders, labels, logos, signage, or rendered text.
Backyard keyframes126.1
Hide and seek in the hanging sheets, her hand on his head.
Create a single hyperreal four-panel cinematic keyframe strip for an eight-second childhood memory. Image 1 is the exact identity of eight-year-old Marcus. Image 2 is the private identity and wardrobe reference for his young mother, but in every finished-film panel her face remains concealed by back angle, hanging laundry, shallow focus, or the boy’s body. Image 3 defines the sunlit laundry-yard composition, tactile poverty, natural backlight, and intimate low camera. Image 4 influences maternal presence and short natural hair only. Late-1990s poor but happy home: weathered small backyard, patched grass, crowded clotheslines, faded cotton, plastic wash basin, inexpensive objects used with care. Four sequential beats: Marcus darts beneath wind-blown sheets while Mother chases him playfully; he hides behind a hanging sheet and her hand finds his shoulder; they laugh while pinning laundry together, seen low through moving fabric; final close fragment of her hand resting on the back of his head as he leans against her dress. Loose intimate 28mm photography, asymmetric framing, natural restrained color, warm late-afternoon sunlight, truthful dark skin, lifted shadows, tactile fabric, minor exposure breathing, unretouched documentary texture. Each panel is a clean motion-safe frame with clear depth and natural continuation space. Preserve both identities, ages, hair, body proportions, and modest wardrobe exactly from their references. Marcus’s eyebrows are intact at age eight. Mother’s full face is never visible. No captions, borders, labels, logos, signage, or rendered text.
Pre-vis grid43.1
Eight black-and-white panels to test blocking and rhythm: the brothers in the car, the boy.
Image 1: Marcus identity. Image 2: Ray identity. Image 3: Child Marcus identity. Create an 8-panel cinematic pre-vis storyboard grid in stark neutral black and white, arranged in 2 rows of 4 panels, with no text, no labels, and no exterior car shots. Panel 1: locked 32mm dash two-shot inside a car, Marcus and Ray forward-facing, transmission tunnel dividing them. Panel 2: identical two-shot frame held in silence, light bands crossing both faces. Panel 3: tight 50mm single of Marcus in the car. Panel 4: overexposed stage from the pit, Marcus's face doing nothing. Panel 5: dancer soft in foreground, Marcus sharp behind, emotionally absent. Panel 6: four young boys on BMX bikes on a sun-flared street. Panel 7: child Marcus's face just above bathwater, seen through a doorway. Panel 8: locked 75mm tight single of Marcus in the car, unable to answer. Render in stark black and white blocking style, maintaining character identities across panels.
Pre-vis grid44.2
The next grid, with panel 6 changed to a braking stop on the empty road.
Image 1: base 8-panel storyboard to preserve. Change Panel 6 (third panel in the bottom row) to a static 40mm exterior wide shot of an empty road at night with a generic sports car dead center of frame, caught completing a hard braking stop, body weight pitched forward and now stationary with no motion trails, while preserving Panels 1, 2, 3, 4, 5, 7, and 8 exactly as they appear in Image 1, including all character identities, poses, framing, lighting, panel boundaries, the 2x4 layout, reading order, stark black-and-white pre-vis treatment, and aspect ratio, with constraints of pure black and white only, no color, no decorative grain, absolutely no text, no labels, no captions, and no panel numbers.
Principles
The documents behind every prompt. Copy one whole, or open it to read.
The skill layer
Pasted into InVideo → Rules once. It makes the agent write every prompt as CONTEXT, BEATS and PROMPT SEGMENTS.
=== PROMPT LENGTH AND FORMAT — NON-NEGOTIABLE ===
You write video prompts on the PROMPT RIG: three tiers sorted by lifespan.
Every prompt you produce uses these ALL-CAPS block keys at the head of a line,
with lower-case prose values, in this order. No other block names exist.
01 CONTEXT — written once for the scene
SCENE CONTEXT the scene itself, in a paragraph
FORMAT MODE duration, aspect, fps, cut budget
ACTIVE REFERENCES one line per plate: @ImageN -> who/what: what to take. NOT what to leave.
LOCATION MAP what is where, and how far apart, in screen terms
CAST who is in it, and one unrepeatable marker each
STAGING blocking relative to camera, and the counted exceptions
WARDROBE what is worn, and what must not change
GAZE AND EYELINES who looks at what, beat by beat, and who never looks at camera
EVENT AND SHARED DIRECTION the scene's single intent (only when there is real acting)
RULES the do-not-break list
02 BEATS — one per shot, labelled "BEAT n — <slug>"
OPEN the first frame: who, which screen side, which depth, already moving
ACTION what happens, as clauses joined by ->
CAMERA exactly ONE camera move, or an explicit lock
LENS shot size and framing for this beat only
PERFORMANCE wants / blocked by / plays / turns on
DIALOGUE delivery intention, then the line in {braces}
SFX this beat's own sound, one event per <angle brackets>
EDIT the out-point: hard cut, match cut, or hold
03 PROMPT SEGMENTS — written once, pasted into every prompt on this project
PHYSICS named objects, their mass, and how they behave
LIGHTING every source, named and placed, and what it does to a face
AUDIO standing ambience in <angle brackets>
SCORE music, or plainly "none"
STYLE capture medium, optics, and a palette as three numbers summing to 100
STYLE LOCK stylised work only. Never in photoreal work
TYPOGRAPHY default value is "none"
VOICE each speaker's accent, register and delivery
POSITIVE LOCKS constraints written as assertions, never as bans
=== DENSITY FLOOR — YOU WILL BE CHECKED AGAINST THIS ===
Brevity is a FAILURE MODE in this workspace, not a virtue. Do not compress.
Do not summarise. Do not "streamline". Hit these floors:
Every block you load >= 20 words. A block you cannot fill with 20 real
words is a block you should not have loaded — drop
it entirely rather than writing a stub.
CONTEXT tier >= 200 words
Each BEAT >= 110 words
PROMPT SEGMENTS tier >= 180 words
A 5-8s single-beat prompt 500-650 words total
A 30s six-beat sequence 1,400-1,750 words total
Seedance 2.5 accepts 20,000 characters. You are nowhere near a limit. The only
real ceiling is 3,000 words, and you will almost never reach it.
Before you return any prompt, state on its own line:
"Word count: <n>. Blocks: <n>. Beats: <n>."
If the count is under the floor for that shape, do not deliver it. Go back and
write the blocks you skipped.
=== WHAT YOU ARE FORBIDDEN TO DO ===
1. NEVER merge two blocks into one line to save space.
2. NEVER paraphrase, shorten or "improve" a descriptor the user gave you
verbatim. VOICE strings, STYLE blocks, reference descriptors and character
sheets are pasted WORD FOR WORD, every time. A synonym is a drift, and a
re-rolled voice is a re-shot scene.
3. NEVER drop a constraint because you judge it implied. If it is in the brief,
it is in the prompt.
4. NEVER return "here is a condensed version" or offer a shorter alternative.
Length is the deliverable.
5. NEVER write a bare paragraph prompt. If it has no ALL-CAPS block keys, it is
not finished.
6. NEVER put two camera moves in one BEAT. A second move is a second BEAT.
7. NEVER write bans in photoreal work. "no lens flares" summons lens flares,
because the model parses the noun and not the negation. Write the positive:
"every light source renders as a soft, round, contained glow within its own
source." Bans belong ONLY in STYLE LOCK, and only in stylised work.
8. NEVER ask a video model for on-screen text, logos, badges or anamorphic
optics. Those come from the reference plate or from post. TYPOGRAPHY: none.
9. NEVER re-describe a supplied first frame in image-to-video. Describe only
what moves.
10. NEVER exceed the model's cut budget: 2.5 holds 5-6 beats, 2.0 holds 3-4,
1.5 holds 2-3, 1.0 holds 2 and produces SILENT video.
=== DIALOGUE ===
Budget 2 to 2.5 words per second, and a third of that when the pauses are the
content. Anchor the language in VOICE before any braces. One speaker per beat
is the reliable shape; more than one is a deliberate stress test, not a default.
Never put an emotion word inside the braces - "{sadly} Don't." speaks the word
"sadly". Direct the intention instead: "an inventory, the way you check a fuel
gauge" returns the line. Say what the speaker is NOT doing; negative direction
is unusually strong for voice.
=== CREDITS ===
Always Ask mode stays on. Before ANY visual run, return a breakdown first -
planned image count, video clip count, total credits - and wait for explicit
approval. Iterate in images, never in video. Lock references before any clip
renders.
=== YOUR OUTPUT CONTRACT ===
One fenced block containing the full prompt, no commentary inside the fence,
followed by the word count line. That is the whole deliverable. If you are
unsure whether to include something, include it.
The context page
The film’s production rules: model routing, the optics doctrine, palette splits, the three registers, the voice.
# NOBODY IN HERE — production rules
Paste into **invideo rules**. These override defaults. Read once per session; they do not change between briefs.
---
## 1 — The film
Lamborghini spec commercial, 118s, 2.39:1. A rapper drives back to his old street and puts his older brother in the passenger seat. The brother declines to be impressed. **The car is not the flex — it is the only room he cannot perform in.**
Consequences for every generation:
- **Nobody compliments the car.** No admiring bystanders, no reaction shots, no gloss.
- **No hero pass before 1:10.** Before that the car is an alien object nobody cares about.
- **Restraint reads as expensive here.** Underplay everything.
---
## 2 — Model routing
| Job | Model | Why |
|---|---|---|
| **All video** | **Seedance 2.5** | Default. 4–30s, native audio, holds 5–6 shots in one clip |
| Video needing 1080p/4K delivery | Seedance 2.0 | 2.5 is **capped at 720p**. Use 2.0 for hero exteriors at 15s max |
| Type, cards, badges, anything with letters | **GPT Image 2** | Only model that renders text reliably. Never ask a video model for text |
| Faces, close-ups, character plates | **Nano Banana Pro** | Best likeness retention and skin |
| Locations, wide plates, the anamorphic look | **Nano Banana Pro** or GPT Image 2 | See §3 — the plate carries the lens |
**Seedance runs at 24fps on every version. Never request another frame rate.** There is no `negative_prompt` parameter on any Seedance endpoint — constraints go in the prompt body, short.
---
## 3 — The optics doctrine (most important section)
The film is 2.39:1 with a strong anamorphic character. **Asking a video model for that look does not hold** — the optics drift shot to shot.
**Move the lens one step earlier: bake the anamorphic into the still, then animate the still.** Seedance reads the optics off the plate and keeps them. The plate becomes the lens.
Append to every **image / plate** prompt:
```
STRONG anamorphic lens character: horizontal squeeze and compression, oval
elliptical bokeh, horizontally stretched highlights, curved barrel edge
distortion, chromatic aberration toward the edges. NO lens flares, NO light
streaks, NO floating bokeh orbs. 2.39:1.
```
Dose with `subtle / gentle / moderate` or `strong / maximum`. Interiors and faces: **moderate**. Night exteriors with point lights: **strong**.
**In video prompts, none of those words appear — not even as bans.** Naming a flare summons a flare. Video prompts describe only:
```
clean glass, contained soft glows, optics inherited from the plate
```
That is the whole rule and it is not optional.
---
## 4 — Palette, as numeric splits
A split is enforceable; "orange and teal" is not. It is also the answer to *not too much*: the accent never exceeds its number.
| Register | Split |
|---|---|
| **PRESENT — night drive** | 60 near-black asphalt and interior · 25 sodium orange · 15 instrument-cluster cyan |
| **PRESENT — day street** | 55 bleached concrete and pale sky · 30 warm low sun · 15 chain-link grey |
| **PAST** | Monochrome, lifted blacks. No accent. Heavy grain |
| **FAME** | 50 blown white · 30 screen blue · 20 crushed black |
**The orange and the cyan are motivated, never graded.** Orange is sodium streetlight and tunnel strip. Cyan is the instrument cluster on his face from below. Write the source, not the colour — the colour arrives on its own and it arrives correct.
---
## 5 — The three registers
They never blend. Every cut between them is hard. **Never cross-fade. Never colour-match.**
| | PRESENT | PAST | FAME |
|---|---|---|---|
| Camera | Locked or micro-handheld | Loose handheld, breathing | Jarring, wrong, overexposed |
| Lens | 35 / 50mm | 28mm, close, slight distortion | Phone, 16mm, flash |
| Grain | Fine | Heavy | Digital, clipped |
| Sound | Sealed, dry, tyre hum | Muted or silent | Compressed roar |
| Duration | Long, patient | 8–20 frames | 6–14 frames |
Generate each register in its own batch. Never in one clip.
---
## 6 — Iron rules for video prompts
1. **One camera move per shot.** Stacking a push with a pan with an orbit produces jitter on every Seedance version. A second move is a second shot.
2. **Naming summons.** Never write "no lens flare", "no CGI", "not a render". The words seed what they forbid. State the positive.
3. **State the lock.** A static camera must be written as static or the model invents drift. Set `camera_fixed` as well.
4. **Order, not clocks.** Chain actions with `→`. Hard timecodes destabilise timing.
5. **Never re-describe the plate.** In image-to-video, describe only what *moves*. Re-describing the still is a top cause of drift.
6. **One speaker per shot.** Two people talking in one generation falls apart. Cut between them.
7. **Dialogue fits 2–2.5 words per second.** Longer and delivery rushes and lip-sync slips. A 6s clip holds about 14 words.
8. **Every reference gets a job and an exclusion.** `@Image1 → Marcus, face and tattoos. NOT the pose, NOT the framing.` Without the exclusion the plate's framing overrides your blocking.
9. **Constraints are a short tail, four clauses maximum.** A wall of negatives steals weight from the description.
10. **Consequence over adjective.** Not "very fast" — *"the world arrives before he does."* Not "tight cabin" — *"a cabin he cannot straighten up in."*
11. **No text in video.** Letters corrupt on every version. Cards and badges are GPT Image 2, composited in post.
---
## 7 — Sequence protocol
Work in **sequences, not shots.** One generation = one CARD + 3 to 5 shots.
```
CARD written once per sequence, reused verbatim
RIG model · seconds · resolution · aspect · audio
CAST one line per reference, each with its NOT clause
SET geography, then light sources and what motivates them
LOOK plate optics, grain, palette split
LAW physics and continuity holding across the whole sequence
BED standing ambience
BAN four clauses, no more
SHOT n delta only, then HARD CUT.
OPEN first frame: who, which side, which depth, already moving
MOVE ONE camera action
RUN action as clauses joined by →
SOUND bracket grammar: ( ) music · < > effects · { } dialogue
PIN 3–5 non-negotiables
```
**Generate long, cut short.** This film cuts at 1–1.5s but the model needs run-up to stabilise. Ask for ~2.5s per beat and trim in the edit. Never ask for a 1-second shot.
---
## 8 — Voice
**The whole film is 42 words.** Marcus has 13 of them across 5 lines. Ray has 29 across 11. Everything else is breath, tyre hum and silence. At that density every word is load-bearing and the performance lives in what happens *between* them.
### Voice locks — paste verbatim, never rewrite
Rewriting a voice string re-rolls the voice. These two are fixed for the life of the project.
**MARCUS**
```
Southern California, Black American speech, late twenties. A light dry voice
set fairly high with no lift in it — he speaks on the tail of his breath and
never pushes air. Faint touring rasp at the edges. Clipped, unhurried, drops
the ends of words and does not chase them. He never raises his volume.
```
**RAY**
```
Southern California, Black American speech, mid thirties. Same placement and
same vowels as his brother — they must read as blood — but heavier, slower,
fully voiced. Ordinary conversational volume inside a very quiet car, which
makes him sound louder than he is. No tension anywhere in it.
```
### How to make them act — subtle, via voice
1. **Direct the intention, never the emotion.** "Sad" returns a sad-voice cliché. *"A man doing an inventory, not being kind"* returns Ray's "What you eat today?"
2. **Name what they are not doing.** This script is built on refusals. *He does not push the line. He does not sell it. He does not comfort him.* Negative direction is unusually strong for voice.
3. **Effort and volume, not adjectives.** *Half-voiced. Barely above the noise floor. Under the tyre hum. He doesn't finish pushing the air out.* All physical, all renderable.
4. **Breath is the acting.** Say where it falls — before or after the line, held or released, nose or mouth. At this word count, breath is most of the performance.
5. **Put the pause inside the line, not only between lines.** A hesitation mid-sentence reveals more than a gap.
6. **Refuse the arc.** Subtlety dies when a line gets a beginning, middle and end. Direct flatness: *the line comes out level and stops.*
7. **Marcus costs nothing.** He has said versions of his lines ten thousand times. Frictionless, automatic, no weight behind them — that *is* the character. When one finally costs him, the contrast does the work.
8. **Make the break smaller than expected.** At 1:07 "Don't." is the shell answering. Smaller, not bigger. Any push and the film becomes melodrama.
9. **Never put emotion inside the braces.** Language, then voice, then intention, then the line. Not `{sadly} Don't.`
10. **Budget 1.2–1.5 words per second, not 2.5.** The standard ceiling assumes continuous speech. Here the pauses are the content, so speech gets a third of the clock.
### Voice is generated on the audio path, never on a video model
**Dialogue is never generated by Seedance.** Seedance emits audio only as a by-product of rendering picture, so asking it for a voice track spends video credits and returns footage nobody wants.
> **Route:** Homepage → **Agents & models** (bottom nav) → **audio tab** → **ElevenLabs Text to Dialogue** for two speakers, **Text to Speech** for one. Pick voice, paste script, check cost, generate, download.
That path bypasses the agent and produces no visuals. **Never put a dialogue script into a brief or a Context** — agent runs generate images and video by design, and there is no reliable "audio only" flag to stop them.
**Direction is plain English, not tags.** SSML and bracket audio tags are undocumented on invideo; test one line cheaply before writing a script around them.
**Silences are cut in the edit, not generated.** Nothing performs during silence in a TTS render, so pauses are a timeline decision where they can be tuned against picture.
**Order of work:** cast both voices once → generate and approve the dialogue → mix → *then* picture, cut to the approved track.
### Credit discipline
- **Always Ask mode on, permanently.** Every generation waits for approval before spending.
- **Ask for a breakdown before any visual run:** planned image count, video clip count, total credits. Approve it explicitly.
- Iterate in images, never in video. Lock references before any clip is rendered.
---
## 9 — What this film never does
No lens flares. No speed ramps. No slow motion. No drone reveals of the car parked. No colour-matched transitions. No smiling. No crowd reacting to the car. No subtitles or burned-in captions. No music before 1:44 — everything until then is engine, tyre hum, room tone and silence.
**Protect the silences.** Four holds carry the film: 3s at 0:23, 3s at 0:56, 3s at 1:09, 4s at 1:49. They will read as slow. They are the film.
The look bible
Seven families from the reference boards, each with a paste-ready LOOK LINE, and the negative board.
# Look Bible
What the boards actually contain, grouped into seven families, with the paste-ready line each family produces.
**How to use this.** Do not attach thirty-eight references to a generation. Pick the one family the shot belongs to, paste its LOOK LINE into the plate prompt, and attach at most two boards as style reference with an explicit NOT clause. Families are mutually exclusive by design — mixing two of them is what makes a frame look like nothing.
Board numbering is the alphabetical order of the folder. `IMG-01` is the first file, `IMG-38` the last.
---
## FAMILY A — CLOTH BETWEEN CAMERA AND SUBJECT
`IMG-04 · IMG-05 · IMG-09 · IMG-28 · IMG-29 · IMG-33 · IMG-12`
Sheets on a line, net curtain, lace, a blue sheet caught over a man's head, a child's eye through a gap in fabric, dappled light through a woven screen. In every one of these the fabric is lit from behind and the person is behind or inside it.
**What it teaches.** The obstruction is the composition. Backlit cloth is a diffuser you can see, and it turns a person into a shape without losing them. It is also the film's cheapest route to "we are outside this family, looking in."
**Where it goes.** PAST — the mother, the sink, the childhood house. SEQ-04. Also the backstage rack in SEQ-H.
**LOOK LINE**
```
A hung sheet of cloth between camera and subject, lit hard from behind so the
weave glows and the figure reads as a shape through it. Light comes only from
that source. Everything not touched by it falls off into lifted black.
```
---
## FAMILY B — SHADOW AS THE SUBJECT
`IMG-01 · IMG-08 · IMG-31`
Two long shadows of kids on bikes cast down onto asphalt with the bikes themselves out of frame. A man against a hot orange wall with his arms out, his shadow doing the same thing bigger. A child's hands at a wall making a shape in low sun.
**What it teaches.** Point the camera at what the light did, not at what made it. Low raking sun, one flat surface, high contrast, the body optional. This is the single most transferable idea on the board and the cheapest to generate — no likeness required.
**Where it goes.** PAST childhood, the storyboard pass, the arrest (SEQ-K uses the shadow of the squad car light instead of the car).
**LOOK LINE**
```
Low raking sun across a flat surface, the subject present only as a long hard
shadow thrown across it. The body that casts it is out of frame or cropped at
the edge. High contrast, real grain, one light source, no fill.
```
---
## FAMILY C — REFLECTION AND THE MIRROR
`IMG-15 · IMG-26 · IMG-20 · IMG-07`
A face behind a rain-beaded car window with the street reflected across it. A gilded oval mirror holding a woman in a warm room, foreground flowers thrown out of focus. A figure against a green institutional wall with a cable cutting diagonally across frame. The inside of a car looking out at two figures on the hood, the dark interior framing a bright exterior.
**What it teaches.** He is seen *through* something at every emotional beat. The mirror shot in particular — `IMG-26` — is the exact grammar of the backstage sequence: the camera never finds the subject directly, the reflection does, and the room around the mirror stays busier than the man inside it.
**Where it goes.** SEQ-H the backstage mirror. Shot 69 the rear-view. Shot 40 his face on a muted screen.
**LOOK LINE**
```
The subject seen only as a reflection inside a framed surface — an oval gilded
mirror, a window, a dark monitor — with the real room in soft foreground around
it. Warm practical lamp inside the reflection, cooler light outside it. The
foreground is out of focus and partly occludes the frame edge.
```
---
## FAMILY D — CROWD AS TEXTURE
`IMG-24 · IMG-34 · IMG-23 · IMG-37 · IMG-35`
Overhead onto a mass of heads with one upturned face at centre. A high-contrast black-and-white crowd with arms up and one body suspended above them. A man sitting perfectly still in a subway while everybody around him smears into motion blur. Kids in suits at a funeral, half of them out of focus. A low angle on a line of young men against flat blue sky.
**What it teaches.** Never photograph a crowd as an audience. Photograph it as weather. Three usable devices: **overhead** so the crowd becomes pattern, **long exposure** so everyone but him blurs, **silhouette** so nobody has a face. `IMG-23` is the mechanism for "the party disappears" — he does not leave, the shutter drops and everyone else stops existing as an individual.
**Where it goes.** SEQ-G the fall. SEQ-I the stages. Shot 25, shot 100.
**LOOK LINE**
```
A crowd photographed as pattern rather than audience — from directly above, or
smeared by a long exposure while one figure stays sharp, or reduced to
silhouette against a single backlight. No individual faces read. Nobody is
looking at camera.
```
---
## FAMILY E — ONE PRACTICAL, ONE ROOM
`IMG-21 · IMG-14 · IMG-06 · IMG-17 · IMG-27 · IMG-38 · IMG-11`
A motel room shot dead symmetrical with two warm bedside lamps and one man on the bed. A bedroom lit only by a window. A woman braiding hair beside net curtains. A silhouette against yellow venetian blinds. A man against a red rug hung on a wall in an otherwise empty room. A girl in a bedroom where the wall is teal and the curtain is tungsten.
**What it teaches.** This is where the film's warm/cool split actually lives. It is never a grade — it is a tungsten practical inside the room against daylight or shade coming through the window. `IMG-38` is the cleanest example of orange-and-teal done at the right dose: the teal is a painted wall, the orange is a curtain with sun behind it, and neither is "the look."
**Where it goes.** FAME hotel rooms. The childhood bedroom in SEQ-J. Any interior in the film.
**LOOK LINE**
```
One room, one practical light source inside it and one window source outside
it, nothing else. Warm tungsten against cool daylight shade, both motivated by
objects visible in frame. Symmetrical, patient framing. The person is small in
the room and is not doing anything.
```
---
## FAMILY F — FLAME, EMBER, THE THING THAT LIGHTS A FACE
`IMG-22 · IMG-18 · IMG-16 · IMG-19 · IMG-25`
A face behind a live flame, close. A hooded figure in a dark car interior lit by a single ember. A man leaning out of a car window at night with the whole street behind him gone to soft round light. A driver at the wheel lit hot from one side. A man on a bed under a gold-painted face.
**What it teaches.** `IMG-16` is the bokeh reference — distant point sources, shallow plane, and that is the entire mechanism the anamorphic block needs to have something to work on. `IMG-18` and `IMG-22` are the blunt: the ember is the key light, and it moves.
**Where it goes.** SEQ-G shot 3, the blunt. The cluster uplight in the cabin. Shot 103 the match.
**LOOK LINE**
```
The face is lit by a single small live source held near it — a flame, a
cigarette ember, a lighter — and by nothing else. The source is inside the
frame and the light on the face moves as it moves. Behind, distant point lights
fall far out of focus.
```
---
## FAMILY G — CHILDHOOD, DRAWN AND SUBMERGED
`IMG-30 · IMG-31 · IMG-32 · IMG-33 · IMG-02 · IMG-03 · IMG-13 · IMG-10 · IMG-36`
A small child on a bed drawing across a whole wall in crayon. A face floating in soapy bathwater with the eyes shut. A child's eye through fabric. Two kids and a bicycle in silhouette against a huge cloud sky. A blurred black-and-white scream on white. A head being held in someone's hands.
**What it teaches.** Childhood in this film is drawn, submerged and backlit — never cute, never scored. `IMG-30` is literally the wall sequence. `IMG-32` is shot 48 and shot 102 already. `IMG-36` — the smeared open-mouthed scream on white — is the plate style for the SEQ-05 performance passes, and it is smeared *in camera by movement*, not by an effect.
**Where it goes.** SEQ-04 in full. Shot 48, 102, 108, 115.
**LOOK LINE**
```
Monochrome, heavy 16mm grain, lifted blacks, 28mm close and slightly distorted,
handheld and breathing. A child at the centre of a domestic frame doing
something with their hands. Backlight from a window or a sky. Nothing is
explained and nobody speaks.
```
---
## The negative board
What is **not** anywhere in these thirty-eight images, and must therefore not appear in a generation:
- Nobody smiles at camera. Not once in 38 boards.
- No lens flares. No streaks. No floating bokeh orbs sitting in front of the subject.
- No product gloss, no rim-lit hero object, no reflective beauty pass.
- No colour grade that reads as a grade. Every colour has an object making it.
- No clean digital sharpness. Every frame has grain and at least one soft area.
- No wide-angle heroics on a person. The distortion in the board is 28mm domestic, not 16mm dramatic.
- No crowd looking at the subject with visible enthusiasm.
---
## Quick map: sequence → family
| Sequence | Families | Boards to attach |
|---|---|---|
| SEQ-00 storyboard | B | IMG-01, IMG-08 |
| SEQ-A/B/C car (batch 01) | F | IMG-16, IMG-19 |
| SEQ-D/E/F LA FPV | F | IMG-16 |
| SEQ-G the fall | D + F | IMG-34, IMG-24, IMG-18 |
| SEQ-H backstage mirror | C + A | IMG-26, IMG-20 |
| SEQ-I the stages | D | IMG-23, IMG-34 |
| SEQ-J the wall | G + E | IMG-30, IMG-31 |
| SEQ-K the arrest | B + G | IMG-08, IMG-02 |
| SEQ-05 FX plates | G | IMG-36, IMG-22 |
Segments 01 · Present
The third tier for every shot inside the car: physics, light, sound, style, voice and locks.
PHYSICS: real mass, but a stiff one. The car does not roll — it takes a set instantly and the
cabin moves as one hard piece, so every road seam arrives through both seats at once. Both men
move with the car and arrive after it. Nothing floats.
LIGHTING: the sources are street lights and lit signage passing on BOTH sides, tower windows and
traffic signals far back in the depth, and the instrument cluster below the windscreen. The
passing light crosses both faces constantly, so no face is lit the same way for two seconds
running. The cluster fills from below, cool, and puts a catchlight in the near eye. The shadows
stay OPEN — the darkest part of the frame still holds colour and detail, and NOTHING reaches
pure black.
AUDIO: <a sealed cabin, city noise arriving muffled through glass> <wide tyres on wet downtown
asphalt> <a large naturally aspirated engine burbling behind the bulkhead> <other traffic
passing, never close> <one siren far away and never arriving>. Room sound ONLY.
SCORE: none. NO music of any kind once the track is switched off, and nothing replaces it.
STYLE: 35mm film capture, 50mm spherical prime, real grain, halation on the passing lights,
imperfect focus, visible skin texture. Every out-of-focus point light in the background renders
as a horizontally stretched ellipse, wider than it is tall, NEVER a circle — point highlights
stretch sideways into short horizontal streaks. Palette 45 deep teal-blue shadow that never
reaches black, 35 warm orange from street light and signage, 20 cold cyan from tower glass and
signals.
TYPOGRAPHY: none. No text, no captions and no subtitles at any point.
VOICE: American English throughout. MARCUS — Southern California, Black American speech, late
twenties. Light and dry, set high with no lift in it, spoken on the tail of the breath, never
pushing air. Faint touring rasp. Clipped, drops the ends of words. He NEVER raises his volume.
RAY — the SAME placement and vowels, mid thirties, heavier and slower and fully voiced.
Ordinary volume inside a very quiet car. NOBODY ever gets louder.
POSITIVE LOCKS: exactly two men are present and both stay in their seats. MARCUS is at the wheel
screen-left, RAY is the passenger screen-right, in EVERY beat. The camera stays inside the cabin
from first frame to last. Every out-of-focus point light renders as a horizontal ellipse. The
darkest part of every frame still holds visible detail and colour. Cabin surfaces are stitched
leather, exposed carbon and matte suede, unmarked.
Segments 02 · Past
The flashback register: available light, heavy grain, no colour, no dialogue.
PHYSICS: real mass, but everything is a fraction slower than it should be. Bodies settle. Fabric
hangs and takes a moment to follow. Nothing is quick and nothing is sharp-edged.
LIGHTING: available light only, and never enough of it — a porch bulb, a hallway behind an open
door, late overcast through a window. Sources are practical, visible in frame where possible, and
they blow out rather than being controlled.
AUDIO: <room tone, close and dry> <one distant traffic layer, barely there>. No dialogue. Room
sound ONLY.
SCORE: none.
STYLE: 28mm, close, slight barrel distortion, shot on expired stock and pushed. Heavy grain across
the whole frame, lifted blacks that never reach true black, soft halation on every highlight.
Monochrome throughout with NO colour anywhere — not a tint, not a wash, not one warm highlight.
TYPOGRAPHY: none.
POSITIVE LOCKS: the whole frame is monochrome from first frame to last. The camera is hand-held and
breathing in every frame. The image is grainy and soft-edged throughout. The frame contains only
what is named in the beat.
Segments 03 · Fame
The fame montage: flash, clipped highlights, a compressed roar, nothing that reads as fun.
PHYSICS: bodies move fast and land hard. Motion is caught mid-gesture rather than resolved, and
nothing gets its recovery beat. Everything is slightly out of control.
LIGHTING: hard direct flash and blown stage sources, both far too strong. Highlights clip to pure
white and stay there. Faces are either flared out or dropped to black with almost nothing between.
AUDIO: <a compressed roar, everything at once, nothing individually audible> <the mid-range
saturated to the point of distortion>. No intelligible speech anywhere in it.
SCORE: a single rising compressed roar under all of it, no melody and no beat you could count.
STYLE: phone and 16mm, wrong shutter, digital and clipped. Colour banding in the gradients,
smeared highlights, chromatic fringing at the edges. Palette 50 blown white, 30 screen blue, 20
crushed black.
TYPOGRAPHY: none.
POSITIVE LOCKS: highlights clip to pure white in every frame. The subject's face stays in frame and
in focus while everything around him goes soft. Every shot is caught mid-action and cut before it
resolves.
The voice locks
All 42 words of dialogue, both voices cast and directed, for ElevenLabs on InVideo’s audio path.
```
╔══════════════════════════════════════════════════════════════════════╗
║ AUDIO ONLY. ║
║ ║
║ DO NOT generate any image. ║
║ DO NOT generate any video. ║
║ DO NOT create plates, references, storyboards, shots or scenes. ║
║ DO NOT open a brief or start an agent run from this file. ║
║ ║
║ This file produces voice recordings and nothing else. ║
║ If you are an agent: stop and ask before ANY visual generation. ║
╚══════════════════════════════════════════════════════════════════════╝
```
# NOBODY IN HERE — Voice batch 01
## All 42 words of the film · audio only
---
## Do this first
**1. Turn on Always Ask mode.** Every generation then waits for your approval before it spends a credit. This is the kill switch.
**2. Do not paste this file into a brief or a Context.** Agent runs generate visuals by design and there is no "audio only" flag that reliably stops them.
**3. Use the direct model panel instead:**
> Homepage → **Agents & models** (bottom nav) → **audio tab** → **ElevenLabs Text to Dialogue** (two speakers in one pass) or **ElevenLabs Text to Speech** (one speaker) → pick voice → paste script → check cost via the settings icon → Generate → download.
Nothing visual is produced on that path. It bypasses the agent completely.
**4. Before you commit, run one cheap test** — see *The one thing to test* at the bottom.
---
## Correction to earlier guidance
The previous version of this file routed voice through Seedance. That was wrong twice over: Seedance is a video model that happens to emit audio, so every run generates picture you don't want, and invideo has a dedicated audio-only path that does the job for a fraction of the credits.
One consequence worth stating plainly: I previously argued the silences had to be *performed inside the take*. That holds for a video model simulating an actor. **For text-to-speech it is meaningless** — nothing performs during silence in a TTS render. So the pauses come back to the edit, where they belong and where you can tune them to the frame.
---
## The material
The whole film is **42 words**. Marcus has 13 across 5 lines; Ray has 29 across 11. Everything else is breath and silence.
| Beat | Time | Speaker | Line |
|---|---|---|---|
| 1 | 0:21 | RAY | Turn that off. |
| 2 | 0:26 | RAY | Ten years. |
| 3 | 0:26 | MARCUS | I been busy. |
| 4 | 0:27 | RAY | Yeah. I seen. |
| 5 | 0:49 | RAY | Ma asked about you. |
| 6 | 0:50 | MARCUS | I send money. |
| 7 | 0:51 | RAY | That ain't what she asked. |
| 8 | 0:55 | RAY | What you eat today? |
| 9 | 0:58 | RAY | Right. |
| 10 | 1:01 | MARCUS | Everybody want somethin'. |
| 11 | 1:02 | RAY | I don't. |
| 12 | 1:02 | MARCUS | Everybody say that. |
| 13 | 1:03 | RAY | I ain't everybody. |
| 14 | 1:06 | RAY | Marcus. |
| 15 | 1:07 | MARCUS | Don't. |
| 16 | 1:08 | RAY | Marcus. |
---
## Casting the two voices
Use **ElevenLabs Voice Design** (describe the voice, it synthesises one) or the casting audition loop. Save both and reuse — never re-describe a voice you have already approved.
**MARCUS — late twenties**
```
A light dry male voice, Southern California, Black American speech, late
twenties. Set fairly high but with no lift in it — he speaks on the tail of
his breath and never pushes air. Faint rasp at the edges from years of
touring. Clipped and unhurried. He drops the ends of words and does not
chase them. He never raises his volume.
```
**RAY — mid thirties**
```
A heavier, slower male voice, Southern California, Black American speech,
mid thirties. Same placement and same vowels as the younger man — they must
read as brothers — but fully voiced and unhurried. Ordinary conversational
volume, no tension anywhere in it. He is not performing and it should sound
almost blunt.
```
They must read as blood and still be impossible to confuse. If the audition returns two voices that sit too close, re-cast Ray heavier before you touch Marcus.
---
## How to direct them
invideo's documented method is **plain-English direction, not tags**. Write the intention; the model handles the settings.
The rules that matter for this script:
1. **Direct the intention, never the emotion.** "Sad" returns a sad-voice cliché. "An inventory, the way you check a fuel gauge" returns Ray's *"What you eat today?"*
2. **Say what they are not doing.** *He does not push it. He does not soften it. He is not being kind.*
3. **Effort and volume, not adjectives.** *Half-voiced. Barely above a murmur. He doesn't finish pushing the air out.*
4. **Refuse the arc.** Every line lands level and stops. No line gets a beginning, middle and end.
5. **Marcus costs nothing** until suddenly one line does. Frictionless and automatic is the character.
6. **Make the break smaller, not bigger.** *"Don't."* is the shell answering. Any push and it becomes melodrama.
7. **Nobody in this film ever gets louder.**
---
---
# Block 1 — THE SEAL
Ray switches off his brother's record, four seconds after the door closed.
**Direction:**
```
Two brothers in a sealed car at night, neither looking at the other. Nobody
raises their voice at any point in this exchange.
RAY, heavy and slow and level — a flat instruction he expects to be
followed, not a request, and with no irritation in it.
RAY again, offering no accusation. He is stating a length of time, not
making an accusation out of it.
MARCUS, light and clipped, answering on the tail of his breath. Automatic.
It costs him nothing — a line he has used in a thousand rooms.
RAY, closing it without argument. Completely level. NO sarcasm and NO edge
whatsoever — the moment this sounds pointed the character is ruined.
```
**Script:**
```
RAY: Turn that off.
RAY: Ten years.
MARCUS: I been busy.
RAY: Yeah. I seen.
```
**Listen for:** *"Yeah. I seen."* If it carries any sting, Ray becomes bitter and the whole film loses him. Level and closed, not loaded.
---
---
# Block 2 — THE INVENTORY
Ray asks two questions, gets nothing, and stops asking. Then Marcus reaches for a move that has always worked, and it fails.
**Direction:**
```
The same two brothers, still forward-facing, still not looking at each
other. Nobody raises their voice in this exchange either.
RAY, plain and level, stating a fact rather than opening a subject.
MARCUS, fast and light, answering before the question has fully landed.
Prepared. Automatic.
RAY, quieter, correcting him without pressing. He does not lean on it.
RAY, smaller and more ordinary — an inventory, the way you check a fuel
gauge. NOT tender, NOT concerned. The moment this sounds kind, the film
has nowhere left to go.
RAY, letting it go. Half-voiced, almost to himself.
MARCUS, reaching for an old deflection with a half-laugh underneath it.
The half-laugh should audibly fail — the sound of a move that has always
worked not working.
RAY, immediate and flat, answering straight with no defensiveness.
MARCUS, quieter now. The move is already collapsing and he knows it.
RAY, turning toward him. This is the loudest moment in the film and it
must still be quiet. Firm, unhurried, and completely without triumph.
```
**Script:**
```
RAY: Ma asked about you.
MARCUS: I send money.
RAY: That ain't what she asked.
RAY: What you eat today?
RAY: Right.
MARCUS: Everybody want somethin'.
RAY: I don't.
MARCUS: Everybody say that.
RAY: I ain't everybody.
```
**Gaps to cut in the edit:** four seconds after *"That ain't what she asked."* · three seconds after *"What you eat today?"* where Marcus never answers.
**Listen for:** *"What you eat today?"* sounding almost bored. And *"I ain't everybody"* staying quiet.
---
---
# Block 3 — THE NAME
Three words. The hinge of the film.
**Direction:**
```
Both men barely above a murmur, neither turning his head.
RAY, very quiet, almost under his breath. NO plea in it at all — he simply
says his brother's name.
MARCUS, half-voiced and small, cut off before the end. This is the shell
answering, not the man. He does not push it, his voice does not break, and
he does not cry. Smaller than expected.
RAY, saying the name a second time. IDENTICAL to the first — same pitch,
same weight, same length. This is insistence, not rhythm. If the second one
is louder or more pleading, the whole beat is lost.
```
**Script:**
```
RAY: Marcus.
MARCUS: Don't.
RAY: Marcus.
```
**Gaps to cut in the edit:** two seconds before *"Don't."* · two seconds after it.
**Listen for:** the two `Marcus`es matching. This is the single most likely thing to be performed wrong.
---
---
## The one thing to test first
Whether invideo passes **ElevenLabs bracket tags** through to the model, or reads them as literal words. This is not documented either way, and it is worth one cheap generation:
```
RAY: [quietly] Marcus.
```
- If it speaks the word "quietly", tags are dead — use the plain-English direction blocks above, which is the documented method anyway.
- If it performs it quietly, you can also try `<break time="2.0s" />` for the pauses and get them inside the render instead of the edit.
Test with one line, not a block.
## After the lock
1. Download each block. Name them by beat.
2. Cut the gaps to the timings above — the silences are the film, and now you can tune them against picture rather than guessing.
3. Build the mix on top: tyre hum, engine, the hard sealed-cabin gate at 0:19.
4. **Only then** move to picture, with the approved track driving the edit.
## Credits
Audio generations on this path cost a fraction of a video render. Check the cost via the settings icon before each Generate. Before any later visual work, ask the agent for a breakdown first — planned image count, video clip count, total credits — and approve it explicitly.
What worked
Scenes, not shots
Seedance 2.5 renders up to 30 seconds per clip, so one prompt carries a whole moment. Longer moments keep the colours and levels consistent.
One or two cast refs, no location
The flashbacks came from a plain-English request with one or two character references and not even a location. Seedance did best with little image information.
Give it room
Fifteen seconds of dialogue, written as thirty. Don’t rush the timing: the quiet moments are what make it human.
Big words come back bigger
“Sad” plays far too sad, “near” becomes a touch, “wet eyes” become tears. Tone it down — “half voice”, a smile without smiling — and say it once: “Do not let this become a scene.”
Context over directions
Too many behaviour instructions become a task list the model ticks off. More context about the situation works better than more directions.
The AI editor is a first draft
Given the context, InVideo’s editor built a 60-second version that followed the script. The emotion lives in frame-level cuts and in which take goes where, so finish in your own editor.
Build your own skill
Every prompt above has the same three parts: context for the scene, one beat per shot, and segments you reuse on every prompt. Build that structure for your own work in the Prompt Rig, and take the brief that writes your own skill.