Back to gallery

Ultimate Cinematic Realism Blueprint

GPT IMAGE 2.5 → NANO BANANA PRO → SEEDANCE 2.5FORMAT 4:3 · 30 S · 1080P · PLATFORM ARCADSPhoto
0 steals

Find an accomplice

Prompt 7

SEEDANCE 2.5 · REFERENCE TO VIDEO · 30 S FORMAT: 30-second cinematic multi-shot sequence, 4:3, 1080p, native dialogue and sound. Hard cuts between shots, real-time action, no slow motion. STYLE: A real night scene filmed on an iPhone, with the texture of an ARRI Alexa and ARRI Signature Prime lenses: low contrast, lifted blacks, soft highlight roll-off, creamy background falloff, muted desaturated grade, fine sensor grain. Handheld operator with natural micro-shake and small breathing drift, no smooth gimbal. Warm amber light from the canopy lanterns and downlights, cool blue-hour sky ambient, wet stone pavement with soft reflections. The penlight is a small hard cool-white source. REFERENCES: @Image 1: LOCATION, BLOCKING AND GRADE ONLY. The hotel entrance under the coffered canopy with hanging brass lanterns, the brass stanchions with black velvet ropes, the round brushed-steel hatch set flush into the wet pavement, the queue receding along the facade, the guard standing to the right of the hatch. Not a first frame: the woman starts one step behind the hatch, not on it. @Image 2: THE WOMAN. Character sheet: face, black layered hair with wispy bangs, septum ring, freckles, oversized white Racing 90 tee over sheer black lace sleeves, black satin slip skirt, red bead necklace, red patent shoulder bag, black Mary Jane flats. Ignore the grid, labels and studio background. @Image 3: THE YOUNG MAN. Character sheet: face, messy black fringe, thin round silver wire glasses, stud earrings, black leather jacket, white shirt, red tie, black wide trousers with a silver chain, chunky black shoes. Ignore the grid, labels and studio background. @Image 4: THE CURLY-HAIRED MAN. Character sheet: face, short dense curly dark brown hair, moustache and goatee with stubble, brown eyes, glossy crinkled burgundy leather bomber jacket worn open, white button-up shirt, cream straight-leg jeans, burgundy suede running sneakers, black square sunglasses. Ignore the grid, labels and studio background. @Image 5: THE GUARD. Character sheet: face, swept-back dark hair, stubble, stern eyes, black button-up shirt, black trousers, black belt, clear coiled earpiece, small black aluminium penlight. Ignore the grid, labels and studio background. @Image 6: THE BOOK. A black linen hardcover; use the front cover design exactly as shown. Use every character sheet to restore each face precisely: same bone structure, skin texture, hair and proportions in every shot. VIDEO PROMPT: 0.0 to 3.0s. Over-the-shoulder from behind the guard, his right shoulder and the back of his head large in the right of frame, earpiece coil visible, exactly the diagonal of @Image 1. The woman stands at the front of the queue, one step behind the steel hatch. She steps forward and plants both feet on the centre of the hatch. The guard lifts the penlight to her eye level, clicks it on and holds the beam on her face. She squints and tilts her head away. The guard speaks, flat and dry. {Are you a real person?} 3.0 to 5.0s. Reverse medium close-up on the woman, eye level, the guard's black shoulder soft in the left foreground. The cool beam hits her eyes and cheek, the rest of her face in warm canopy light. She answers with an offended little smile. {Of course.} 5.0 to 6.5s. Close-up on the guard, slightly low angle, the penlight glare blooming in the lower foreground. He does not blink. {AI.} 6.5 to 9.0s. Close-up on the woman, she leans in, mouth open in outrage, hand flicking at the light, bag strap sliding on her shoulder. {Oh my God, are you kidding me? I am real.} 9.0 to 10.0s. Close-up on the guard, deadpan, one small shake of the head. {Nope.} 10.0 to 11.5s. Top-down shot straight above the hatch. The two steel half-doors split along the centre seam and swing down with a heavy mechanical clunk. The woman drops straight down out of view into darkness, unhurt, her bag following. 11.5 to 13.5s. Camera inside the dark shaft looking straight up through the open round hatch: a bright circle of canopy light, lantern glow and wet rim. The young man from @Image 3 leans over the edge and peers down, glasses catching the light, faces of the queue behind him. The guard's voice comes from above. {Next.} The half-doors swing up and slam shut, frame to black. 13.5 to 16.0s. Over-the-shoulder behind the guard again, same diagonal as the opening. The young man steps onto the hatch, pushes his glasses up, chin raised, cool. The beam hits his face and reflects as two small dots in his lenses. {I am real, man.} 16.0 to 18.0s. Close-up on the guard, flat. {Nope.} Cut wide: the half-doors drop, the young man vanishes straight down, tie flicking upward, the doors snap shut. Behind the rope the curly-haired man, sunglasses on, tilts his chin slightly, unbothered. 18.0 to 20.0s. Medium shot from the guard's side. The curly-haired man from @Image 4 strolls forward wearing the black sunglasses. As his foot lands on the hatch he takes the sunglasses off with his right hand in one smooth unhurried move and folds them, a small half-smile. The guard raises the penlight to his bare eyes. He answers calmly. {Real as it gets.} 20.0 to 22.0s. Extreme macro close-up of the curly-haired man's right eye filling the frame: brown iris fibres, wet tear film, individual lashes, fine skin texture at the edge. The penlight beam enters and the pupil visibly constricts in a quick smooth reflex, the bright point reflecting on the cornea. The beam slides away and the pupil slowly widens back. Fine skin texture, real moisture. 22.0 to 23.5s. Medium close-up on the guard. He clicks the light off, gives one short nod and unhooks the velvet rope with his free hand. {All right. You are in.} 23.5 to 26.5s. Tracking from behind, low shoulder height: the curly-haired man walks past the guard toward the glass doors, easy rolling stride, burgundy leather catching the lantern light, folded sunglasses in his right hand. Over his right shoulder we look down at his left hand: he carries the black hardcover book from @Image 6, cover facing up, swinging gently with each step. Lantern light slides across the cloth. 26.5 to 30.0s. Close-up on the book in his hand as he walks, camera matching his pace, then he slows and the cover holds steady and sharp to the end. The cover reads exactly, in large white condensed letters: ULTIMATE CINEMATIC REALISM BLUEPRINT, with the arcads logo underneath and the film strip of four portraits below, exactly as on @Image 6. Final 1.5 seconds hold on the cover. AUDIO: Diegetic only, no music. City night ambience under everything: distant traffic, soft crowd murmur from the queue, a faint drip from the canopy edge, footsteps on wet stone. <penlight click> at 1.5s and again at 22.0s. <heavy steel latch clunk and hydraulic hiss> at 10.0s and 16.5s. A short surprised yelp from each falling character fading downward. <hatch slam echoing in the shaft> at 13.0s. <velvet rope hook unclipping> at 22.8s. Soft cloth rustle of the book against his hand from 24.0s. VOICES: GUARD: male, mid forties, low flat American English voice, dry, clipped, quiet authority, never shouting. WOMAN: female, early twenties, American English, slightly husky alt-girl voice, offended and rising into disbelief. YOUNG MAN: male, about twenty, soft light tenor, American English with a subtle Korean accent, cool and a little mumbled. CURLY-HAIRED MAN: male, mid-thirties, natural American English accent, smooth warm baritone with a slight husky grain, relaxed and effortlessly confident, a subtle smile in the voice, clear articulation, never shouting. Each line is spoken once, by its own character, in the shot where it appears. CONSISTENCY: Every face matches its character sheet exactly in every shot. The guard keeps the earpiece and the black penlight in his right hand for the whole video; the penlight never leaves his hand. The young man never removes his round wire glasses. The curly-haired man wears the sunglasses until he stands on the hatch, removes them there and never puts them back on. The woman keeps the red bag on her shoulder until she drops. Queue order never changes: woman, young man, curly-haired man, then strangers. The hatch is the same round steel hatch in every shot. Location, lanterns and low-contrast grade stay identical. The book cover is identical to @Image 6. NEGATIVE PROMPT: cars, high contrast, crushed blacks, oversaturated color, smooth gimbal, slow motion, character sheet grids or labels, subtitles, captions, extra spoken words, narration of the prompt, garbled book title, changed faces, young man's glasses removed, teal cap, tattoos on the curly-haired man, sunglasses back on after removal, extra people on the hatch, injury, blood, watermark

Other prompts in this tutorial

Prompt 1

GPT IMAGE 2.5 · CHARACTER CHART · THE WOMAN { "task": "Photorealistic character reference sheet for AI film production, assembled from real photographs taken in one studio session.", "references": { "image_1": "WARDROBE ONLY. Oversized white tee with the red 'Racing '90' print and black graphic, sheer black floral lace long-sleeve top underneath, black satin ankle-length slip skirt, red bead necklace, thin silver ring, small red patent leather shoulder bag, thick black square-frame glasses. Copy the exact shape, print, and finish. NEVER use this woman's face, hair, or body.", "image_2": "IDENTITY ONLY. Face, bone structure, brown eyes, freckles, skin tone, glossy rosy lips, septum ring, black shoulder-length layered hair with wispy bangs. Hood removed in all panels.", "image_3": "LAYOUT ONLY. Grid structure, light grey seamless background, panel proportions, uppercase sans-serif labels. Do NOT copy the man, the burgundy leather jacket, the sneakers, or the original label words." }, "layout": { "top_row": [ "Full-body front, neutral stance, red bag held in left hand, neutral expression, eyes to lens", "Full-body left profile, wearing the black square glasses", "Full-body back view, plain tee back, lace sleeves visible", "Large head-and-shoulders close-up, front, no glasses, necklace visible", "Three-quarter close-up wearing the black square glasses, slight head turn, eyes just past camera" ], "bottom_row_labels_exact": ["BLACK HAIR / BANGS", "LACE SLEEVES", "GRAPHIC TEE", "PATENT BAG", "SKIN TEXTURE / PIERCING", "BROWN EYES", "COLOR PALETTE"], "palette_swatches": ["optic white", "cherry red", "black", "charcoal lace", "rose lip pink", "espresso black-brown"] }, "wardrobe_addition": "Black patent leather Mary Jane flats with a thin strap.", "camera": "Full-body panels at 85mm f/8, close-ups at 100mm f/5.6, eye-level, identical white balance across panels.", "lighting": "Large soft octabox key from front-left, gentle fill, soft contact shadows under feet, light grey seamless paper with a faint floor curve.", "realism": "Unretouched skin with visible pores, peach fuzz, uneven freckles, slight redness around the nose. A few flyaway hairs. Real cotton creases and slight print cracking. Skin visible through the lace mesh. Soft satin folds. Softbox reflection on the patent bag. Use case: professional casting and wardrobe board.", "constraints": "Same woman and same outfit in every panel. No beauty retouching, no plastic skin, no perfect symmetry, no extra people, no watermark, no text other than the seven labels and the tee print." }

Prompt 2

GPT IMAGE 2.5 · CHARACTER CHART · THE YOUNG MAN { "task": "Photorealistic character reference sheet for AI film production, assembled from real photographs taken in one studio session.", "references": { "image_1": "IDENTITY ONLY. The young man's face, bone structure, eye shape, skin tone, full soft lips, and his short black hair with a messy textured fringe falling onto the forehead. His own thin silver round wire-frame glasses are part of his identity.", "image_2": "WARDROBE ONLY (neck down). Boxy black leather jacket with snap-button collar and zip, crisp white button-up shirt worn untucked at the hem, the long crimson red silk tie with the white logo near the tip, black trousers with silver chain hardware at the waist, silver rings on several fingers. Copy the exact shape, finish, and details. Ignore all hands, the green drink, the phone, the laptop, and the flying papers.", "image_3": "ACCESSORY ONLY. The silver wraparound sport sunglasses with chrome mirrored lenses, and the small silver stud earrings. Ignore the face around them.", "image_4": "LAYOUT ONLY. Grid structure, light grey seamless background, panel proportions, uppercase sans-serif labels. Do NOT copy the woman, the lace top, the red bag, the red necklace, or the original label words." }, "layout": { "top_row": [ "Full-body front, neutral stance, arms relaxed at sides, jacket open showing the shirt and tie, wearing his round wire glasses, neutral expression, eyes to lens", "Full-body left profile, wearing the chrome wraparound sunglasses", "Full-body back view, back of the leather jacket and his black hair visible", "Large head-and-shoulders close-up, front, wearing his round wire glasses, earrings and tie knot visible", "Three-quarter close-up wearing the chrome wraparound sunglasses, slight head turn, chin slightly raised" ], "bottom_row_labels_exact": ["BLACK HAIR", "LEATHER JACKET", "RED TIE", "WRAPAROUND SHADES", "RINGS / EARRINGS", "SKIN TEXTURE", "COLOR PALETTE"], "palette_swatches": ["jet black", "optic white", "crimson red", "chrome silver", "charcoal", "warm skin beige"] }, "wardrobe_addition": "Black leather chunky-sole derby shoes.", "camera": "Full-body panels at 85mm f/8, close-ups at 100mm f/5.6, eye-level, identical white balance across panels.", "lighting": "Large soft octabox key from front-left, gentle fill, soft contact shadows under feet, light grey seamless paper with a faint floor curve.", "realism": "Unretouched skin with visible pores, slight unevenness and a few tiny marks on the cheeks, faint shadow of stubble on the upper lip and chin, natural lip texture. Individual flyaway strands in the fringe. Leather with creases at the elbows, soft scuffs, and a satin sheen. Cotton shirt wrinkles. Silk tie with a subtle woven texture. Chrome reflections in the sunglasses showing the softbox. Use case: professional casting and wardrobe board.", "constraints": "Same young man and same outfit in every panel. His face comes only from Image 1. No beauty retouching, no plastic skin, no perfect symmetry, no extra people, no extra hands, no watermark, no text other than the seven labels and the tie logo." }

Prompt 3

GPT IMAGE 2.5 · CHARACTER CHART · THE GUARD { "task": "Photorealistic character reference sheet for AI film production, assembled from real photographs taken in one studio session. The character is a private security guard at a venue entrance who checks people's faces with a small flashlight.", "references": { "image_1": "IDENTITY ONLY. The man's face, lean angular bone structure, deep horizontal forehead lines, crow's feet, dark brown eyes under a heavy stern brow, short dark stubble with a few grey hairs on the chin, dark hair swept back with receding temples, prominent neck and Adam's apple. Keep his stern, unimpressed expression.", "image_2": "LAYOUT ONLY. Grid structure, light grey seamless background, panel proportions, uppercase sans-serif labels. Do NOT copy the woman, her clothes, her bag, her necklace, or the original label words." }, "wardrobe": "Plain black long-sleeve button-up shirt in a matte cotton twill, top button open, sleeves down and buttoned at the cuff, neatly tucked in. Straight-leg black work trousers with a sharp crease. Plain black leather belt with a dull steel buckle. Black leather lace-up duty shoes with thick rubber soles. A clear coiled acoustic-tube earpiece in his right ear, running down behind the collar. A small matte black aluminum tactical penlight. No logos, no badges, no printed text on the clothing.", "layout": { "top_row": [ "Full-body front, feet shoulder-width apart, hands clasped in front at belt level holding the penlight, stern neutral expression, eyes to lens", "Full-body left profile, right arm raised forward at chest height pointing the switched-on penlight ahead as if checking someone's face, a thin bright beam", "Full-body back view, earpiece tube visible behind the collar, shirt tucked in", "Large head-and-shoulders close-up, front, stern unimpressed stare straight into the lens, earpiece visible", "Three-quarter close-up holding the switched-on penlight up near his cheek, beam pointing past the camera, cool white spill light grazing his jaw and stubble, eyes narrowed in scrutiny" ], "bottom_row_labels_exact": ["SWEPT-BACK HAIR", "BLACK SHIRT", "PENLIGHT", "EARPIECE", "BELT / TROUSERS", "SKIN TEXTURE / STUBBLE", "COLOR PALETTE"], "palette_swatches": ["matte black", "charcoal", "graphite", "steel grey", "weathered skin tan", "cool LED white"] }, "camera": "Full-body panels at 85mm f/8, close-ups at 100mm f/5.6, eye-level, identical white balance across panels.", "lighting": "Large soft octabox key from front-left, gentle fill, soft contact shadows under feet, light grey seamless paper with a faint floor curve. In the penlight panels, the beam adds a small hard cool-white highlight without changing the overall studio light.", "realism": "Unretouched weathered skin with visible pores, deep forehead creases, fine wrinkles around the eyes, slight redness on the nose, uneven stubble density with grey hairs on the chin. A few loose strands in the swept-back hair. Matte black fabric that still shows folds, stitching, and faint lint. Creases in the trousers at the knee. Scuffs on the shoe toes. Fine machining marks and wear on the penlight body. Use case: professional casting and wardrobe board.", "constraints": "Same man and same outfit in every panel. His face comes only from Image 1. No beauty retouching, no plastic skin, no perfect symmetry, no weapons, no extra people, no watermark, no text other than the seven labels." }

Prompt 4

NANO BANANA PRO · THE SET Image 1 is the location: the hotel entrance with the cream stone facade, the coffered canopy with warm downlights and hanging brass lanterns, the glass entrance doors, the wet pavement, and the blue-hour city skyline. Keep this architecture and mood, but remove the black car, the luggage carts, and the staff standing at the doors. The curb in front is empty. Images 2 to 5 are character reference sheets. From each one, take only the person's face, hair, and outfit. Ignore the grid layout, the grey background, and every label. Image 5 is the security guard: black button-up shirt, black trousers, a clear coiled earpiece, and a small black penlight. Image 2 is the young woman with black bangs and a septum ring, in the oversized white "Racing '90" tee over sheer black lace sleeves, the black satin slip skirt, the red bead necklace, and the red patent shoulder bag. Image 3 is the young man with the round wire glasses, the black leather jacket, the white shirt, and the red tie. Image 4 is the curly-haired bearded man with tattoo sleeves, the cream graphic tee, the teal cap, the washed wide-leg jeans, and the tan suede boots. Scene: blue hour. A queue of people waits along the facade beneath the canopy, held in a lane by polished brass stanchions with black velvet ropes. At the very front of the queue, set flush into the stone pavement, is a round brushed-steel hatch about one and a half metres wide, with a thin polished steel rim and a faint seam across its centre. It is a trapdoor, closed for now. The woman from Image 2 stands on the hatch. The security guard from Image 5 stands just in front of her, seen from behind over his right shoulder at a slight diagonal: we see the back of his head, his ear with the earpiece, part of his jaw in profile, and his raised right arm aiming the switched-on penlight straight into her face. She squints and leans back slightly in the hard white beam. Next in line, waiting a step behind the rope, is the young man from Image 3, watching with a neutral face. Behind him stands the bearded man from Image 4. Behind them the queue continues deep into the frame along the building: thirty or more ordinary strangers in dark coats and casual night-out clothes, of varied ages, chatting and checking phones, softly out of focus. Camera: a real photograph taken on an iPhone, held just behind the guard's shoulder at chest height, vertical frame. The image carries the texture of an ARRI Alexa with ARRI Signature Prime lenses: gentle low-contrast rendering, soft highlight roll-off, lifted blacks, creamy falloff into the background, with the queue receding diagonally toward the vanishing point. Light: warm amber glow from the canopy downlights and lanterns, cool blue ambient from the sky, and the penlight's small hard cool-white beam as the brightest point, on her face. Wet pavement with soft reflections of the lanterns and the steel rim of the hatch. Realism: unretouched skin with natural pores, real fabric wrinkles, slight phone sensor noise in the shadows, a muted desaturated grade. No car, no captions, no signage lettering, no watermark, and nothing copied from the character sheet labels.

Prompt 5

NANO BANANA PRO · EDIT · THE SWAP Edit Image 1. Keep everything in Image 1 exactly the same: the camera angle, framing, location, canopy and lanterns, the wet pavement, the brass stanchions and velvet ropes, the steel hatch, the guard with the penlight, the woman on the hatch, the young man with the glasses and red tie, every stranger in the long queue, the lighting, the low-contrast grade, and the grain. Change only one person: the third person in the queue, the man in the teal cap and cream graphic tee standing directly behind the young man with the red tie. Replace him completely with the man from Image 2, a character reference sheet. From Image 2 take only his face, his short dense curly dark brown hair, his moustache and goatee with stubble, and his outfit: the glossy crinkled burgundy leather bomber jacket worn open, the white button-up shirt, the cream straight-leg jeans, and the burgundy suede running sneakers. He wears the black square sunglasses from Image 2. Ignore the grid, the labels, and the studio background of Image 2. He stands in exactly the same spot, at the same scale and height, in the same relaxed standing pose, waiting in line behind the young man and partly overlapped by him in the same way, head turned slightly toward the hatch. He is lit by the same warm amber canopy light and cool blue ambient as everyone around him, with matching softness, grain, and color, so he sits naturally in the photograph. The glossy leather picks up small warm highlights from the lanterns. No cap, no tattoos, no other changes anywhere in the image.

Prompt 6

GPT IMAGE 2.5 · THE BOOK { "task": "Premium product photography of one hardcover book, shown from four angles in a 2x2 grid of equal panels separated by thin white gutters, pure white seamless background in every panel. Use case: hero visual for a launch video.", "references": { "image_1": "LOGO. The Arcads logo: angular mark plus the lowercase wordmark 'arcads'. Reproduce its exact shapes and proportions, printed in white. Do not redraw or restyle it.", "image_2": "CHARACTER SHEET: use only the young woman's face (black bangs, septum ring). Ignore grid and labels.", "image_3": "CHARACTER SHEET: use only the young man's face (round wire glasses, messy black fringe). Ignore grid and labels.", "image_4": "CHARACTER SHEET: use only the bearded man's face (wavy hair, freckles). Ignore grid and labels.", "image_5": "CHARACTER SHEET: use only the security guard's face (swept-back hair, stubble, stern stare). Ignore grid and labels." }, "book": { "format": "Large-format hardcover, roughly 24 x 32 cm, 4 cm thick, rounded spine, black headbands, a thin crimson silk ribbon bookmark, page edges sprayed matte black.", "material": "Soft-touch matte black linen-cloth binding with a fine visible weave. Title in crisp bright white foil, very slightly debossed into the cloth. The four portraits printed with a gloss spot-UV finish that catches the light against the matte cloth.", "cover_design": "Strict Swiss editorial grid, generous margins. Top left, small: 'VOL. 01'. Top right, small: 'FIELD MANUAL'. Upper half: the title set huge, left-aligned, all caps, stacked on four lines exactly: 'ULTIMATE' / 'CINEMATIC' / 'REALISM' / 'BLUEPRINT', in a tall, tightly kerned condensed neo-grotesk display face (Druk-style), filling the full width. Directly under the title, the Arcads logo from Image 1 in white, medium size. Lower third: a horizontal 35mm film-strip band running edge to edge, four frames with sprocket holes, each frame a tight cinematic close-up portrait of one character in this order: the guard from Image 5 lit by a small penlight glare, the woman from Image 2, the young man from Image 3, the bearded man from Image 4. Portraits in a muted, low-contrast cinematic color grade with fine grain. Tiny frame numbers under each frame: '01', '02', '03', '04'. Bottom edge, small, letter-spaced caps: 'A FIELD MANUAL FOR PHOTOREAL AI FILMMAKING'.", "spine": "Title running vertically in the same condensed white face, Arcads angular mark at the base, 'VOL. 01' at the top.", "back_cover": "Mostly empty matte black. One short white line centered in the upper third: 'EVERY FRAME IS A DECISION.' A small white barcode block bottom right, the Arcads logo small bottom left." }, "panels": [ "Top left: the front cover photographed straight-on, perfectly flat and centered, filling most of the panel.", "Top right: the book standing upright at a three-quarter angle, showing the front cover and the full spine.", "Bottom left: the back cover, straight-on, with the spine edge just visible.", "Bottom right: the book lying flat at a low 30-degree angle, slightly open, showing the black sprayed page edges and the crimson ribbon trailing out onto the white surface." ], "photography": "Shot on a 100mm lens at f/11, everything tack sharp. Large soft overhead softbox plus a white bounce, soft natural contact shadows under the book on the white sweep. Real material behavior: the matte cloth absorbs light and shows its weave, the white foil has a slight metallic sheen, the spot-UV portraits show a soft gloss reflection, the corners are crisp.", "constraints": "Identical book design in all four panels. All text spelled exactly as given, no other text anywhere. Logo exactly as in Image 1. No hands, no props, no watermark." }

Tutorial

DENIZ AKKABAK · ULTIMATE REALISM TUTORIAL · @DENIZ.AKKABAK

MADE IN ARCADS · arcads.ai

The finished 30-second film — four characters, one trapdoor, one book. (Open this document in Google Docs to watch it play.)
The finished 30-second film — four characters, one trapdoor, one book. (Open this document in Google Docs to watch it play.)

Ultimate Cinematic Realism Blueprint

Cast from Pinterest, lock four characters into studio charts, build the set, design the hero prop and direct a 30-second door-check film with native dialogue — every step made in Arcads.

A TUTORIAL BY DENIZ AKKABAK · @DENIZ.AKKABAK

PIPELINE GPT IMAGE 2.5 → NANO BANANA PRO → SEEDANCE 2.5 FORMAT 4:3 · 30 S · 1080P PLATFORM ARCADS

Every viral door check runs on one question: are you a real person? This film turns that question on the viewer. A bouncer shines a penlight into each face, a steel trapdoor swallows anyone who fails, and the joke only lands if the audience genuinely cannot tell who is synthetic. That makes realism the punchline — not a nice-to-have. This blueprint walks through the entire build exactly as we made it in Arcads: casting outfits from Pinterest, locking four characters into studio-grade charts, compositing the set, designing the hero book and directing a single 30-second Seedance 2.5 generation with four distinct voices. Every prompt is printed verbatim, and every “why it works” is tied to the exact line that does the work.

THE PIPELINE AT A GLANCE

  1. 01

    Cast from Pinterest

    Pair one face with one outfit per character. The face never borrows the outfit’s body, and the outfit never borrows the face.

  2. 02

    Lock the charts

    GPT Image 2.5 turns each pair into a five-angle studio sheet with labelled detail tiles — the continuity bible every later model restores from.

  3. 03

    Build the set

    Nano Banana Pro drops the cast into the real location; a surgical edit then swaps me into the queue without moving a single pixel of the rest.

  4. 04

    Design the prop

    GPT Image 2.5 renders the hero book in four angles with the title, cover portraits and Arcads logo intact.

  5. 05

    Shoot the film

    Seedance 2.5 reads six references and directs fourteen shots, four voices and a working trapdoor in one 30-second generation.

THE MODEL MAP — WHICH MODEL FOR WHICH JOB

  1. 01

    GPT Image 2.5

    Charts, typography and anything with text. It obeys reference scopes literally and spells labels, prints and logos cleanly. We ran 16:9 for charts and 9:16 for the book, 2K, quality high.

  2. 02

    Nano Banana Pro

    Location compositing and keep-everything-change-one-thing edits. The richest material texture at 4K, with the aspect ratio matched to the frame you need.

  3. 03

    Nano Banana 2.1

    The fast draft lane. Useful for blocking a composition cheaply, but in our side-by-side test its faces drifted toward the outfit references — never trust it with final identity.

  4. 04

    Seedance 2.5

    The director. Reference to Video mode, up to 30 seconds, multi-shot editing, native dialogue and sound. We rendered 30 s, 4:3, 1080p, audio on, MOV, high bitrate.

STEP 01 — PINTEREST → ARCADS

Casting From Pinterest

Realism starts before the first prompt. For each character we pulled two images: a face, which owns identity, and a Pinterest outfit, which owns wardrobe. Keeping them in separate images is the whole trick — a model can only blend what you hand it inside one picture, so the face reference decides bone structure, skin and hair, while the outfit reference decides cut, fabric and print and its model’s face is explicitly banned. The guard needed no outfit photo at all: plain black workwear is easier to write than to find. For the third character I skipped casting entirely and used my own chart.

REFERENCE INPUTS — ONE FACE, ONE OUTFIT

Face — the woman
Face — the woman
Pinterest outfit — Racing ’90 tee, lace sleeves, satin skirt
Pinterest outfit — Racing ’90 tee, lace sleeves, satin skirt
Face — the young man
Face — the young man
Pinterest outfit — cropped below the chin (see Step 02)
Pinterest outfit — cropped below the chin (see Step 02)
Face — the guard; his wardrobe is written, not referenced
Face — the guard; his wardrobe is written, not referenced
My own chart — I play the third character
My own chart — I play the third character

STEP 02 — GPT IMAGE 2.5

Locking the Cast — Character Charts

Before committing, we generated the same chart side by side with Nano Banana 2.1, Nano Banana Pro and GPT Image 2.5. GPT Image 2.5 won clearly: the closest identity, the cleanest labels and the fewest leaks. Nano Banana Pro pulled a stranger’s eyes into a detail tile, and Nano Banana 2.1 drifted the faces toward the outfit models. Every chart uses the same three-slot architecture — WARDROBE ONLY, IDENTITY ONLY, LAYOUT ONLY — each with a NEVER clause that names what it must not contribute. The layout slot is simply any finished chart: it teaches the grid, the grey seamless and the label typography without lending a face.

Chart 1 — the woman

REFERENCE INPUTS — ATTACH IN THIS ORDER

Image 1 — the Pinterest outfit. Image 2 — the face. Image 3 — a finished chart as the layout template (we used my own).

Prompt 1

GPT IMAGE 2.5 · CHARACTER CHART · THE WOMAN { "task": "Photorealistic character reference sheet for AI film production, assembled from real photographs taken in one studio session.", "references": { "image_1": "WARDROBE ONLY. Oversized white tee with the red 'Racing '90' print and black graphic, sheer black floral lace long-sleeve top underneath, black satin ankle-length slip skirt, red bead necklace, thin silver ring, small red patent leather shoulder bag, thick black square-frame glasses. Copy the exact shape, print, and finish. NEVER use this woman's face, hair, or body.", "image_2": "IDENTITY ONLY. Face, bone structure, brown eyes, freckles, skin tone, glossy rosy lips, septum ring, black shoulder-length layered hair with wispy bangs. Hood removed in all panels.", "image_3": "LAYOUT ONLY. Grid structure, light grey seamless background, panel proportions, uppercase sans-serif labels. Do NOT copy the man, the burgundy leather jacket, the sneakers, or the original label words." }, "layout": { "top_row": [ "Full-body front, neutral stance, red bag held in left hand, neutral expression, eyes to lens", "Full-body left profile, wearing the black square glasses", "Full-body back view, plain tee back, lace sleeves visible", "Large head-and-shoulders close-up, front, no glasses, necklace visible", "Three-quarter close-up wearing the black square glasses, slight head turn, eyes just past camera" ], "bottom_row_labels_exact": ["BLACK HAIR / BANGS", "LACE SLEEVES", "GRAPHIC TEE", "PATENT BAG", "SKIN TEXTURE / PIERCING", "BROWN EYES", "COLOR PALETTE"], "palette_swatches": ["optic white", "cherry red", "black", "charcoal lace", "rose lip pink", "espresso black-brown"] }, "wardrobe_addition": "Black patent leather Mary Jane flats with a thin strap.", "camera": "Full-body panels at 85mm f/8, close-ups at 100mm f/5.6, eye-level, identical white balance across panels.", "lighting": "Large soft octabox key from front-left, gentle fill, soft contact shadows under feet, light grey seamless paper with a faint floor curve.", "realism": "Unretouched skin with visible pores, peach fuzz, uneven freckles, slight redness around the nose. A few flyaway hairs. Real cotton creases and slight print cracking. Skin visible through the lace mesh. Soft satin folds. Softbox reflection on the patent bag. Use case: professional casting and wardrobe board.", "constraints": "Same woman and same outfit in every panel. No beauty retouching, no plastic skin, no perfect symmetry, no extra people, no watermark, no text other than the seven labels and the tee print." }

OUTPUT

GPT Image 2.5 · 16:9 · 2K · quality high — five angles, seven detail tiles, one identity.
GPT Image 2.5 · 16:9 · 2K · quality high — five angles, seven detail tiles, one identity.

Chart 2 — the young man, and the crop trick

Our first male charts kept inheriting the outfit model’s face: the model reads a person in a photo as a package, so the jaw and moustache rode along with the jacket. The fix happens before the prompt. Crop the outfit photo below the chin, and cut a second, tight crop for the one accessory that sits on the face. The outfit model’s face never enters the generation, so it cannot leak. Note how the prompt then tells the model what to ignore inside the crops — hands, drinks, phones, papers.

REFERENCE INPUTS — ATTACH IN THIS ORDER

Image 1 — the face. Image 2 — outfit, neck down. Image 3 — accessory band only. Image 4 — the woman’s chart as layout.

Image 2 — outfit, cropped below the chin
Image 2 — outfit, cropped below the chin
Image 3 — sunglasses and earrings band only
Image 3 — sunglasses and earrings band only

Prompt 2

GPT IMAGE 2.5 · CHARACTER CHART · THE YOUNG MAN { "task": "Photorealistic character reference sheet for AI film production, assembled from real photographs taken in one studio session.", "references": { "image_1": "IDENTITY ONLY. The young man's face, bone structure, eye shape, skin tone, full soft lips, and his short black hair with a messy textured fringe falling onto the forehead. His own thin silver round wire-frame glasses are part of his identity.", "image_2": "WARDROBE ONLY (neck down). Boxy black leather jacket with snap-button collar and zip, crisp white button-up shirt worn untucked at the hem, the long crimson red silk tie with the white logo near the tip, black trousers with silver chain hardware at the waist, silver rings on several fingers. Copy the exact shape, finish, and details. Ignore all hands, the green drink, the phone, the laptop, and the flying papers.", "image_3": "ACCESSORY ONLY. The silver wraparound sport sunglasses with chrome mirrored lenses, and the small silver stud earrings. Ignore the face around them.", "image_4": "LAYOUT ONLY. Grid structure, light grey seamless background, panel proportions, uppercase sans-serif labels. Do NOT copy the woman, the lace top, the red bag, the red necklace, or the original label words." }, "layout": { "top_row": [ "Full-body front, neutral stance, arms relaxed at sides, jacket open showing the shirt and tie, wearing his round wire glasses, neutral expression, eyes to lens", "Full-body left profile, wearing the chrome wraparound sunglasses", "Full-body back view, back of the leather jacket and his black hair visible", "Large head-and-shoulders close-up, front, wearing his round wire glasses, earrings and tie knot visible", "Three-quarter close-up wearing the chrome wraparound sunglasses, slight head turn, chin slightly raised" ], "bottom_row_labels_exact": ["BLACK HAIR", "LEATHER JACKET", "RED TIE", "WRAPAROUND SHADES", "RINGS / EARRINGS", "SKIN TEXTURE", "COLOR PALETTE"], "palette_swatches": ["jet black", "optic white", "crimson red", "chrome silver", "charcoal", "warm skin beige"] }, "wardrobe_addition": "Black leather chunky-sole derby shoes.", "camera": "Full-body panels at 85mm f/8, close-ups at 100mm f/5.6, eye-level, identical white balance across panels.", "lighting": "Large soft octabox key from front-left, gentle fill, soft contact shadows under feet, light grey seamless paper with a faint floor curve.", "realism": "Unretouched skin with visible pores, slight unevenness and a few tiny marks on the cheeks, faint shadow of stubble on the upper lip and chin, natural lip texture. Individual flyaway strands in the fringe. Leather with creases at the elbows, soft scuffs, and a satin sheen. Cotton shirt wrinkles. Silk tie with a subtle woven texture. Chrome reflections in the sunglasses showing the softbox. Use case: professional casting and wardrobe board.", "constraints": "Same young man and same outfit in every panel. His face comes only from Image 1. No beauty retouching, no plastic skin, no perfect symmetry, no extra people, no extra hands, no watermark, no text other than the seven labels and the tie logo." }

OUTPUT

His face comes only from Image 1 — round wire glasses as identity, chrome shades as wardrobe.
His face comes only from Image 1 — round wire glasses as identity, chrome shades as wardrobe.

Chart 3 — the guard

No outfit photo here. When the wardrobe is generic, writing it is more precise than finding it: matte twill, a sharp trouser crease, a dull steel buckle, a clear coiled earpiece, a machined penlight. The prompt also gives him his job — checking faces with a small flashlight — so two panels rehearse the exact gesture the film needs.

REFERENCE INPUTS — ATTACH IN THIS ORDER

Image 1 — the face. Image 2 — the woman’s chart as layout.

Prompt 3

GPT IMAGE 2.5 · CHARACTER CHART · THE GUARD { "task": "Photorealistic character reference sheet for AI film production, assembled from real photographs taken in one studio session. The character is a private security guard at a venue entrance who checks people's faces with a small flashlight.", "references": { "image_1": "IDENTITY ONLY. The man's face, lean angular bone structure, deep horizontal forehead lines, crow's feet, dark brown eyes under a heavy stern brow, short dark stubble with a few grey hairs on the chin, dark hair swept back with receding temples, prominent neck and Adam's apple. Keep his stern, unimpressed expression.", "image_2": "LAYOUT ONLY. Grid structure, light grey seamless background, panel proportions, uppercase sans-serif labels. Do NOT copy the woman, her clothes, her bag, her necklace, or the original label words." }, "wardrobe": "Plain black long-sleeve button-up shirt in a matte cotton twill, top button open, sleeves down and buttoned at the cuff, neatly tucked in. Straight-leg black work trousers with a sharp crease. Plain black leather belt with a dull steel buckle. Black leather lace-up duty shoes with thick rubber soles. A clear coiled acoustic-tube earpiece in his right ear, running down behind the collar. A small matte black aluminum tactical penlight. No logos, no badges, no printed text on the clothing.", "layout": { "top_row": [ "Full-body front, feet shoulder-width apart, hands clasped in front at belt level holding the penlight, stern neutral expression, eyes to lens", "Full-body left profile, right arm raised forward at chest height pointing the switched-on penlight ahead as if checking someone's face, a thin bright beam", "Full-body back view, earpiece tube visible behind the collar, shirt tucked in", "Large head-and-shoulders close-up, front, stern unimpressed stare straight into the lens, earpiece visible", "Three-quarter close-up holding the switched-on penlight up near his cheek, beam pointing past the camera, cool white spill light grazing his jaw and stubble, eyes narrowed in scrutiny" ], "bottom_row_labels_exact": ["SWEPT-BACK HAIR", "BLACK SHIRT", "PENLIGHT", "EARPIECE", "BELT / TROUSERS", "SKIN TEXTURE / STUBBLE", "COLOR PALETTE"], "palette_swatches": ["matte black", "charcoal", "graphite", "steel grey", "weathered skin tan", "cool LED white"] }, "camera": "Full-body panels at 85mm f/8, close-ups at 100mm f/5.6, eye-level, identical white balance across panels.", "lighting": "Large soft octabox key from front-left, gentle fill, soft contact shadows under feet, light grey seamless paper with a faint floor curve. In the penlight panels, the beam adds a small hard cool-white highlight without changing the overall studio light.", "realism": "Unretouched weathered skin with visible pores, deep forehead creases, fine wrinkles around the eyes, slight redness on the nose, uneven stubble density with grey hairs on the chin. A few loose strands in the swept-back hair. Matte black fabric that still shows folds, stitching, and faint lint. Creases in the trousers at the knee. Scuffs on the shoe toes. Fine machining marks and wear on the penlight body. Use case: professional casting and wardrobe board.", "constraints": "Same man and same outfit in every panel. His face comes only from Image 1. No beauty retouching, no plastic skin, no perfect symmetry, no weapons, no extra people, no watermark, no text other than the seven labels." }

OUTPUT

The penlight panels are a rehearsal: Seedance later restores this exact grip and beam.
The penlight panels are a rehearsal: Seedance later restores this exact grip and beam.

STEP 03 — NANO BANANA PRO

Building the Set

The set is a real hotel entrance photo plus the four charts. Two decisions matter. First, the location is told what to lose — the car, the luggage carts, the doormen — so the frame belongs to the queue. Second, the trapdoor is written as an engineered object with a size, a polished rim and a centre seam, because the film will open it later and the model needs to know where it splits. The camera lines carry the realism anchors that run through the whole project: a real photograph taken on an iPhone, with the texture of ARRI Signature Prime glass and a deliberately low-contrast grade.

REFERENCE INPUTS — ATTACH IN THIS ORDER

Image 1 — the location
Image 1 — the location
Images 2–5 — the four charts (the third slot held a stand-in we later replaced)
Images 2–5 — the four charts (the third slot held a stand-in we later replaced)

Prompt 4

NANO BANANA PRO · THE SET Image 1 is the location: the hotel entrance with the cream stone facade, the coffered canopy with warm downlights and hanging brass lanterns, the glass entrance doors, the wet pavement, and the blue-hour city skyline. Keep this architecture and mood, but remove the black car, the luggage carts, and the staff standing at the doors. The curb in front is empty. Images 2 to 5 are character reference sheets. From each one, take only the person's face, hair, and outfit. Ignore the grid layout, the grey background, and every label. Image 5 is the security guard: black button-up shirt, black trousers, a clear coiled earpiece, and a small black penlight. Image 2 is the young woman with black bangs and a septum ring, in the oversized white "Racing '90" tee over sheer black lace sleeves, the black satin slip skirt, the red bead necklace, and the red patent shoulder bag. Image 3 is the young man with the round wire glasses, the black leather jacket, the white shirt, and the red tie. Image 4 is the curly-haired bearded man with tattoo sleeves, the cream graphic tee, the teal cap, the washed wide-leg jeans, and the tan suede boots. Scene: blue hour. A queue of people waits along the facade beneath the canopy, held in a lane by polished brass stanchions with black velvet ropes. At the very front of the queue, set flush into the stone pavement, is a round brushed-steel hatch about one and a half metres wide, with a thin polished steel rim and a faint seam across its centre. It is a trapdoor, closed for now. The woman from Image 2 stands on the hatch. The security guard from Image 5 stands just in front of her, seen from behind over his right shoulder at a slight diagonal: we see the back of his head, his ear with the earpiece, part of his jaw in profile, and his raised right arm aiming the switched-on penlight straight into her face. She squints and leans back slightly in the hard white beam. Next in line, waiting a step behind the rope, is the young man from Image 3, watching with a neutral face. Behind him stands the bearded man from Image 4. Behind them the queue continues deep into the frame along the building: thirty or more ordinary strangers in dark coats and casual night-out clothes, of varied ages, chatting and checking phones, softly out of focus. Camera: a real photograph taken on an iPhone, held just behind the guard's shoulder at chest height, vertical frame. The image carries the texture of an ARRI Alexa with ARRI Signature Prime lenses: gentle low-contrast rendering, soft highlight roll-off, lifted blacks, creamy falloff into the background, with the queue receding diagonally toward the vanishing point. Light: warm amber glow from the canopy downlights and lanterns, cool blue ambient from the sky, and the penlight's small hard cool-white beam as the brightest point, on her face. Wet pavement with soft reflections of the lanterns and the steel rim of the hatch. Realism: unretouched skin with natural pores, real fabric wrinkles, slight phone sensor noise in the shadows, a muted desaturated grade. No car, no captions, no signage lettering, no watermark, and nothing copied from the character sheet labels.

OUTPUT

Nano Banana Pro · 4K · reframed to 5:4 in a quick edit pass that locked the queue order.
Nano Banana Pro · 4K · reframed to 5:4 in a quick edit pass that locked the queue order.

The swap — putting myself in line

Nano Banana Pro is the strongest model we have for “keep everything, change one thing”. The prompt first lists everything that must survive — angle, lanterns, stanchions, hatch, every stranger, the grain — and only then names the single change, down to the spot, the scale, the overlap with the person in front and the light that must fall on the new leather. The sunglasses are on deliberately: the film opens with me hidden behind them.

REFERENCE INPUTS — ATTACH IN THIS ORDER

Image 1 — the set
Image 1 — the set
Image 2 — my chart
Image 2 — my chart

Prompt 5

NANO BANANA PRO · EDIT · THE SWAP Edit Image 1. Keep everything in Image 1 exactly the same: the camera angle, framing, location, canopy and lanterns, the wet pavement, the brass stanchions and velvet ropes, the steel hatch, the guard with the penlight, the woman on the hatch, the young man with the glasses and red tie, every stranger in the long queue, the lighting, the low-contrast grade, and the grain. Change only one person: the third person in the queue, the man in the teal cap and cream graphic tee standing directly behind the young man with the red tie. Replace him completely with the man from Image 2, a character reference sheet. From Image 2 take only his face, his short dense curly dark brown hair, his moustache and goatee with stubble, and his outfit: the glossy crinkled burgundy leather bomber jacket worn open, the white button-up shirt, the cream straight-leg jeans, and the burgundy suede running sneakers. He wears the black square sunglasses from Image 2. Ignore the grid, the labels, and the studio background of Image 2. He stands in exactly the same spot, at the same scale and height, in the same relaxed standing pose, waiting in line behind the young man and partly overlapped by him in the same way, head turned slightly toward the hatch. He is lit by the same warm amber canopy light and cool blue ambient as everyone around him, with matching softness, grain, and color, so he sits naturally in the photograph. The glossy leather picks up small warm highlights from the lanterns. No cap, no tattoos, no other changes anywhere in the image.

OUTPUT

Nano Banana Pro · 5:4 · 4K — this frame becomes @Image 1 for the film.
Nano Banana Pro · 5:4 · 4K — this frame becomes @Image 1 for the film.

STEP 04 — GPT IMAGE 2.5

Designing the Hero Prop

The ending pays off on an object, so the object has to be worth the close-up. We designed the book with the cast on its cover and rendered it from four angles on white, so the video model has every face of the prop — front, spine, back and page edges. GPT Image 2.5 is the right engine because the cover is mostly typography: four stacked lines in a condensed display face, small caps, frame numbers and a logo that must not be redrawn. The material language — soft-touch linen, white foil, spot-UV portraits, a crimson ribbon — is what makes it read as a real object rather than a mockup.

REFERENCE INPUTS — ATTACH IN THIS ORDER

Image 1 — the Arcads logo. Images 2–5 — the character charts, used for faces only.

Prompt 6

GPT IMAGE 2.5 · THE BOOK { "task": "Premium product photography of one hardcover book, shown from four angles in a 2x2 grid of equal panels separated by thin white gutters, pure white seamless background in every panel. Use case: hero visual for a launch video.", "references": { "image_1": "LOGO. The Arcads logo: angular mark plus the lowercase wordmark 'arcads'. Reproduce its exact shapes and proportions, printed in white. Do not redraw or restyle it.", "image_2": "CHARACTER SHEET: use only the young woman's face (black bangs, septum ring). Ignore grid and labels.", "image_3": "CHARACTER SHEET: use only the young man's face (round wire glasses, messy black fringe). Ignore grid and labels.", "image_4": "CHARACTER SHEET: use only the bearded man's face (wavy hair, freckles). Ignore grid and labels.", "image_5": "CHARACTER SHEET: use only the security guard's face (swept-back hair, stubble, stern stare). Ignore grid and labels." }, "book": { "format": "Large-format hardcover, roughly 24 x 32 cm, 4 cm thick, rounded spine, black headbands, a thin crimson silk ribbon bookmark, page edges sprayed matte black.", "material": "Soft-touch matte black linen-cloth binding with a fine visible weave. Title in crisp bright white foil, very slightly debossed into the cloth. The four portraits printed with a gloss spot-UV finish that catches the light against the matte cloth.", "cover_design": "Strict Swiss editorial grid, generous margins. Top left, small: 'VOL. 01'. Top right, small: 'FIELD MANUAL'. Upper half: the title set huge, left-aligned, all caps, stacked on four lines exactly: 'ULTIMATE' / 'CINEMATIC' / 'REALISM' / 'BLUEPRINT', in a tall, tightly kerned condensed neo-grotesk display face (Druk-style), filling the full width. Directly under the title, the Arcads logo from Image 1 in white, medium size. Lower third: a horizontal 35mm film-strip band running edge to edge, four frames with sprocket holes, each frame a tight cinematic close-up portrait of one character in this order: the guard from Image 5 lit by a small penlight glare, the woman from Image 2, the young man from Image 3, the bearded man from Image 4. Portraits in a muted, low-contrast cinematic color grade with fine grain. Tiny frame numbers under each frame: '01', '02', '03', '04'. Bottom edge, small, letter-spaced caps: 'A FIELD MANUAL FOR PHOTOREAL AI FILMMAKING'.", "spine": "Title running vertically in the same condensed white face, Arcads angular mark at the base, 'VOL. 01' at the top.", "back_cover": "Mostly empty matte black. One short white line centered in the upper third: 'EVERY FRAME IS A DECISION.' A small white barcode block bottom right, the Arcads logo small bottom left." }, "panels": [ "Top left: the front cover photographed straight-on, perfectly flat and centered, filling most of the panel.", "Top right: the book standing upright at a three-quarter angle, showing the front cover and the full spine.", "Bottom left: the back cover, straight-on, with the spine edge just visible.", "Bottom right: the book lying flat at a low 30-degree angle, slightly open, showing the black sprayed page edges and the crimson ribbon trailing out onto the white surface." ], "photography": "Shot on a 100mm lens at f/11, everything tack sharp. Large soft overhead softbox plus a white bounce, soft natural contact shadows under the book on the white sweep. Real material behavior: the matte cloth absorbs light and shows its weave, the white foil has a slight metallic sheen, the spot-UV portraits show a soft gloss reflection, the corners are crisp.", "constraints": "Identical book design in all four panels. All text spelled exactly as given, no other text anywhere. Logo exactly as in Image 1. No hands, no props, no watermark." }

OUTPUT

GPT Image 2.5 · 9:16 · 2K · quality high — every line of type spelled correctly on the first pass.
GPT Image 2.5 · 9:16 · 2K · quality high — every line of type spelled correctly on the first pass.

STEP 05 — SEEDANCE 2.5

Directing the Film

One generation, thirty seconds, fourteen shots. Seedance 2.5 in Reference to Video mode reads six images and treats each one exactly as scoped: the set for blocking and grade, four charts for restoring faces, the book for the final hold. The woman starts one step behind the hatch rather than on it, so the film opens with movement instead of a frozen still. Each beat is a timecode, a physical camera position and an end state; every line of dialogue appears once, inside braces, with its own voice described in the VOICES block. Everything here runs in the same Arcads studio — start in Arcads and attach the references in the order below.

REFERENCE INPUTS — ATTACH IN THIS ORDER

@Image 1 — set, blocking, grade
@Image 1 — set, blocking, grade
@Image 2 — the woman
@Image 2 — the woman
@Image 3 — the young man
@Image 3 — the young man
@Image 4 — me
@Image 4 — me
@Image 5 — the guard
@Image 5 — the guard
@Image 6 — the book
@Image 6 — the book

SETTINGS

Seedance 2.5 — Reference to Video · 30 s · 4:3 · 1080p · native audio on · MOV · high bitrate

Prompt 7

SEEDANCE 2.5 · REFERENCE TO VIDEO · 30 S

30-second cinematic multi-shot sequence, 4:3, 1080p, native dialogue and sound. Hard cuts between shots, real-time action, no slow motion.

A real night scene filmed on an iPhone, with the texture of an ARRI Alexa and ARRI Signature Prime lenses: low contrast, lifted blacks, soft highlight roll-off, creamy background falloff, muted desaturated grade, fine sensor grain. Handheld operator with natural micro-shake and small breathing drift, no smooth gimbal. Warm amber light from the canopy lanterns and downlights, cool blue-hour sky ambient, wet stone pavement with soft reflections. The penlight is a small hard cool-white source.

@Image 1: LOCATION, BLOCKING AND GRADE ONLY. The hotel entrance under the coffered canopy with hanging brass lanterns, the brass stanchions with black velvet ropes, the round brushed-steel hatch set flush into the wet pavement, the queue receding along the facade, the guard standing to the right of the hatch. Not a first frame: the woman starts one step behind the hatch, not on it. @Image 2: THE WOMAN. Character sheet: face, black layered hair with wispy bangs, septum ring, freckles, oversized white Racing 90 tee over sheer black lace sleeves, black satin slip skirt, red bead necklace, red patent shoulder bag, black Mary Jane flats. Ignore the grid, labels and studio background. @Image 3: THE YOUNG MAN. Character sheet: face, messy black fringe, thin round silver wire glasses, stud earrings, black leather jacket, white shirt, red tie, black wide trousers with a silver chain, chunky black shoes. Ignore the grid, labels and studio background. @Image 4: THE CURLY-HAIRED MAN. Character sheet: face, short dense curly dark brown hair, moustache and goatee with stubble, brown eyes, glossy crinkled burgundy leather bomber jacket worn open, white button-up shirt, cream straight-leg jeans, burgundy suede running sneakers, black square sunglasses. Ignore the grid, labels and studio background. @Image 5: THE GUARD. Character sheet: face, swept-back dark hair, stubble, stern eyes, black button-up shirt, black trousers, black belt, clear coiled earpiece, small black aluminium penlight. Ignore the grid, labels and studio background. @Image 6: THE BOOK. A black linen hardcover; use the front cover design exactly as shown. Use every character sheet to restore each face precisely: same bone structure, skin texture, hair and proportions in every shot.

0.0 to 3.0s. Over-the-shoulder from behind the guard, his right shoulder and the back of his head large in the right of frame, earpiece coil visible, exactly the diagonal of @Image 1. The woman stands at the front of the queue, one step behind the steel hatch. She steps forward and plants both feet on the centre of the hatch. The guard lifts the penlight to her eye level, clicks it on and holds the beam on her face. She squints and tilts her head away. The guard speaks, flat and dry. {Are you a real person?} 3.0 to 5.0s. Reverse medium close-up on the woman, eye level, the guard's black shoulder soft in the left foreground. The cool beam hits her eyes and cheek, the rest of her face in warm canopy light. She answers with an offended little smile. {Of course.} 5.0 to 6.5s. Close-up on the guard, slightly low angle, the penlight glare blooming in the lower foreground. He does not blink. {AI.} 6.5 to 9.0s. Close-up on the woman, she leans in, mouth open in outrage, hand flicking at the light, bag strap sliding on her shoulder. {Oh my God, are you kidding me? I am real.} 9.0 to 10.0s. Close-up on the guard, deadpan, one small shake of the head. {Nope.} 10.0 to 11.5s. Top-down shot straight above the hatch. The two steel half-doors split along the centre seam and swing down with a heavy mechanical clunk. The woman drops straight down out of view into darkness, unhurt, her bag following. 11.5 to 13.5s. Camera inside the dark shaft looking straight up through the open round hatch: a bright circle of canopy light, lantern glow and wet rim. The young man from @Image 3 leans over the edge and peers down, glasses catching the light, faces of the queue behind him. The guard's voice comes from above. {Next.} The half-doors swing up and slam shut, frame to black. 13.5 to 16.0s. Over-the-shoulder behind the guard again, same diagonal as the opening. The young man steps onto the hatch, pushes his glasses up, chin raised, cool. The beam hits his face and reflects as two small dots in his lenses. {I am real, man.} 16.0 to 18.0s. Close-up on the guard, flat. {Nope.} Cut wide: the half-doors drop, the young man vanishes straight down, tie flicking upward, the doors snap shut. Behind the rope the curly-haired man, sunglasses on, tilts his chin slightly, unbothered. 18.0 to 20.0s. Medium shot from the guard's side. The curly-haired man from @Image 4 strolls forward wearing the black sunglasses. As his foot lands on the hatch he takes the sunglasses off with his right hand in one smooth unhurried move and folds them, a small half-smile. The guard raises the penlight to his bare eyes. He answers calmly. {Real as it gets.} 20.0 to 22.0s. Extreme macro close-up of the curly-haired man's right eye filling the frame: brown iris fibres, wet tear film, individual lashes, fine skin texture at the edge. The penlight beam enters and the pupil visibly constricts in a quick smooth reflex, the bright point reflecting on the cornea. The beam slides away and the pupil slowly widens back. Fine skin texture, real moisture. 22.0 to 23.5s. Medium close-up on the guard. He clicks the light off, gives one short nod and unhooks the velvet rope with his free hand. {All right. You are in.} 23.5 to 26.5s. Tracking from behind, low shoulder height: the curly-haired man walks past the guard toward the glass doors, easy rolling stride, burgundy leather catching the lantern light, folded sunglasses in his right hand. Over his right shoulder we look down at his left hand: he carries the black hardcover book from @Image 6, cover facing up, swinging gently with each step. Lantern light slides across the cloth. 26.5 to 30.0s. Close-up on the book in his hand as he walks, camera matching his pace, then he slows and the cover holds steady and sharp to the end. The cover reads exactly, in large white condensed letters: ULTIMATE CINEMATIC REALISM BLUEPRINT, with the arcads logo underneath and the film strip of four portraits below, exactly as on @Image 6. Final 1.5 seconds hold on the cover.

Diegetic only, no music. City night ambience under everything: distant traffic, soft crowd murmur from the queue, a faint drip from the canopy edge, footsteps on wet stone. <penlight click> at 1.5s and again at 22.0s. <heavy steel latch clunk and hydraulic hiss> at 10.0s and 16.5s. A short surprised yelp from each falling character fading downward. <hatch slam echoing in the shaft> at 13.0s. <velvet rope hook unclipping> at 22.8s. Soft cloth rustle of the book against his hand from 24.0s.

male, mid forties, low flat American English voice, dry, clipped, quiet authority, never shouting.

female, early twenties, American English, slightly husky alt-girl voice, offended and rising into disbelief.

male, about twenty, soft light tenor, American English with a subtle Korean accent, cool and a little mumbled.

male, mid-thirties, natural American English accent, smooth warm baritone with a slight husky grain, relaxed and effortlessly confident, a subtle smile in the voice, clear articulation, never shouting. Each line is spoken once, by its own character, in the shot where it appears.

Every face matches its character sheet exactly in every shot. The guard keeps the earpiece and the black penlight in his right hand for the whole video; the penlight never leaves his hand. The young man never removes his round wire glasses. The curly-haired man wears the sunglasses until he stands on the hatch, removes them there and never puts them back on. The woman keeps the red bag on her shoulder until she drops. Queue order never changes: woman, young man, curly-haired man, then strangers. The hatch is the same round steel hatch in every shot. Location, lanterns and low-contrast grade stay identical. The book cover is identical to @Image 6.

cars, high contrast, crushed blacks, oversaturated color, smooth gimbal, slow motion, character sheet grids or labels, subtitles, captions, extra spoken words, narration of the prompt, garbled book title, changed faces, young man's glasses removed, teal cap, tattoos on the curly-haired man, sunglasses back on after removal, extra people on the hatch, injury, blood, watermark

OUTPUT — FOUR CHECKPOINTS

The squint under the beam · the top-down drop · the view from inside the shaft · the pupil reflex.
The squint under the beam · the top-down drop · the view from inside the shaft · the pupil reflex.

STEP 06

Anatomy — Why It Reads as Real

  1. 01

    The contract

    “30-second cinematic multi-shot sequence … Hard cuts between shots, real-time action, no slow motion.” The first line tells Seedance it is cutting an edit, not drifting through one shot, and “real-time” removes the floaty slow-motion that gives AI footage away.

  2. 02

    Hybrid realism

    “filmed on an iPhone, with the texture of an ARRI Alexa and ARRI Signature Prime lenses: low contrast, lifted blacks …” The phone supplies believability — handheld micro-shake, sensor grain, “no smooth gimbal” — while the cinema glass supplies falloff. Low contrast reads as footage, not a render.

  3. 03

    Scoped references

    “LOCATION, BLOCKING AND GRADE ONLY … Not a first frame … Ignore the grid, labels and studio background.” Every slot states what it controls and what it must not. Without this, the model opens on the still or prints chart labels into the street.

  4. 04

    Restore, don’t describe

    “Use every character sheet to restore each face precisely.” The charts are the source of truth; the short text descriptions are a second lock. Upstream, the crop trick guarantees there is nothing foreign in a face to restore.

  5. 05

    Angles that prove the set

    “Camera inside the dark shaft looking straight up through the open round hatch …” Fake sets only work from the front. A top-down drop and a POV from inside the pit show the space from angles a viewer has never seen — the strongest evidence that the place exists.

  6. 06

    Physiological truth

    “the pupil visibly constricts in a quick smooth reflex … slowly widens back.” Real pupils tighten under light and relax after it. Getting the biology right in an extreme macro is the moment the audience stops doubting.

  7. 07

    Sound, voices, insurance

    “<heavy steel latch clunk and hydraulic hiss> … Each line is spoken once … extra spoken words, narration of the prompt, garbled book title” Timecoded SFX give every cut a physical cause, the VOICES block keeps four speakers distinct, and the negative prompt blocks the classic failures before they happen.

THE RESULT

Output — the Door Check

Seedance 2.5 · 30 s · 4:3 · 1080p with native dialogue. (Open this document in Google Docs to watch it play.)
Seedance 2.5 · 30 s · 4:3 · 1080p with native dialogue. (Open this document in Google Docs to watch it play.)

Audit it against the prompt: the woman steps onto the steel before the beam hits her, the shaft POV catches the next in line peering down, my sunglasses come off the moment my foot lands and never return, the pupil tightens and relaxes, and the final hold reads the cover cleanly. If the title ever wobbles, planar-track the GPT Image 2.5 book render over the hold.

THE BOUNCER ASKS THE QUESTION YOUR AUDIENCE IS ALREADY ASKING

If the bouncer can’t tell, neither can the feed.

Swap the hotel for a club, the penlight for a scanner, the trapdoor for a sliding wall — the skeleton carries any test-and-reveal format. Cast from real references, lock every face into a chart, build the set as a still, design the prop that pays off the ending, then let one long reference-to-video generation direct the edit. Realism isn’t a filter you add at the end; it is a chain of small, scoped decisions, and every link in this blueprint is one of them.

Every model in this blueprint — GPT Image 2.5, Nano Banana Pro and Seedance 2.5 — lives in one studio. Cast your own characters, build your set and direct your first 30-second film in Arcads.

arcads.ai →

WRITTEN, DIRECTED & PROMPTED BY DENIZ AKKABAK · @DENIZ.AKKABAK