비 갠 파리 뒷골목, 전화하며 걷던 여자가 지나가던 사람의 팔꿈치에 스칩니다. 목걸이 줄이 끊어지고 커피컵과 접이식 휴대폰이 같은 순간 손을 떠나는데, 거기서 시간이 3퍼센트 속도로 늘어납니다.
이 작품을 굴리는 힘은 하나입니다. 물건은 제자리에 있고 카메라만 움직인다는 규칙입니다. 화면에서 목걸이가 커지는 것은 목걸이가 렌즈로 다가와서가 아니라 카메라가 거리를 좁혔기 때문입니다. 이 한 줄을 지키려고 물건마다 화면 좌우 위치를 못 박고, 접사를 끝낸 뒤 처음의 넓은 화면으로 똑같은 각도로 돌아오게 했어요. 아무것도 움직이지 않았다는 것을 말로 주장하는 대신 화면이 스스로 보여주게 만든 장치입니다.
레퍼런스는 세 장만 썼습니다. 인물 캐릭터 시트 한 장, 파리 거리 한 장, 크리스탈 스노우글로브 목걸이 제품 시트 한 장입니다. 커피컵과 접이식 휴대폰은 일부러 레퍼런스를 만들지 않았어요. 화면에 오래 남고 정체성이 흔들리면 안 되는 것만 이미지로 고정하고 나머지는 글로 잡는 편이, 주인공 얼굴에 쓸 자리를 안 깎기 때문입니다.
목걸이를 주인공 소품으로 고른 이유도 그 규칙에서 나왔습니다. 양손이 폰과 커피로 차 있으니 세 번째 물건은 몸에 걸려 있어야 부딪힘만으로 떨어져 나갑니다. 그리고 시간이 멈춘 영상에서, 안에 든 금박이 영영 가라앉지 않는 스노우글로브는 그냥 예쁜 소품이 아니라 이 영상이 왜 시간을 멈췄는지에 대한 대답이 됩니다. 드레스는 짜낸 크림처럼 보이지만 조각한 옷감이라고 못 박아, 넘어질 때도 뭉개지지 않게 했어요.
단계 4개 · 레퍼런스 3장
따라 만들기
각 단계의 프롬프트를 복사해 이미지를 만들고, 이미지마다 붙은 @Image 번호 순서 그대로 업로드하면 마지막 영상 프롬프트의 번호가 정확히 맞습니다.
Step 1
케이크 드레스 인물 캐릭터 시트
케이크 드레스 인물 캐릭터 시트
1번째로 업로드하면 프롬프트의 @Image 1이 됩니다
생성 프롬프트 · 3583자
Character reference sheet as FOUR full-body views in a row on a clean white studio background — FRONT, 3/4, SIDE PROFILE, BACK — plus one large head-and-shoulders face close-up inset panel. Identity strictly from the attached reference face.
Subject: a woman with warm chestnut-auburn hair, long and waved past the shoulders, the front sections braided back into a half-up crown with loose curled tendrils framing the face; dewy skin with a soft flush across the cheeks and the bridge of the nose, long lashes, glossy rose lips, calm neutral expression.
Outfit — a strapless couture mini dress sculpted to look like piped buttercream. The fitted bustier is pale butter-yellow, its surface built from broad horizontal palette-knife frosting strokes in butter-yellow and cream, with three large sculpted rose swirls across the chest and a smooth cream frosting band at the waist. From the waist falls a very full round bubble-hem mini skirt, its entire surface covered edge to edge in large piped butter-yellow rose swirls, with a few soft pink piped pinwheel blossoms set into the yellow at the hip and along the hem. A pair of glossy red cherries on a single green stem is pinned at the center front waist beside two tiny gold pins. The bodice back repeats the same horizontal frosting strokes and closes under a soft pink piped ribbon bow at the center back waist; the rose swirls wrap continuously around the back of the skirt with no flat or unfinished panel.
Legs and feet — bare legs, no socks and no stockings. Ivory satin ribbon about two centimetres wide rises out of the topline of each shoe: it crosses over the front of the ankle, wraps behind, crosses again up the shin, and ties in a flat knot at the back of the calf just below the knee, leaving two short ribbon tails hanging. Four evenly spaced crossings per leg, the ribbon lying flat against the skin, its satin catching a soft highlight along each edge and dimpling slightly where it draws snug. In the FRONT and 3/4 views the crossings read as clean diagonal lines climbing the shin; in the SIDE PROFILE the wrap is seen edge-on; in the BACK view the knot and both hanging tails are clearly visible. On her feet, cream satin round-toe mary-jane shoes with a low block heel, a small pearl button on the strap and a fine piped-frosting trim along the topline.
Worn accessories: sheer white floral-lace gloves on both hands, ending at the wrist with a ruffled cuff; one large round white pearl drop on each ear; nothing at the neck.
Both hands are completely empty and relaxed at her sides — she holds nothing in any view.
Same person, same outfit, same neutral standing pose logic in every view, rotated only by the stated angle. Tall runway-model proportions, roughly nine heads tall — small head, long legs carrying more than half of standing height — and every full-body view fills its panel from crown to shoe, never squashed or shrunk down to fit the frame.
The attached photo supplies facial identity and the garment design only; do not carry over its bakery interior, its seated pose on the counter, its camera height, its held object or its window light — every view follows the plain standing lookbook turnaround described above.
Clean even studio lighting like a fashion lookbook turnaround, true color and honest garment weight: the soft matte grain of piped frosting, the sheerness of the lace, the satin sheen of the ribbon and shoes. Real skin texture with visible pores and natural catchlights in the eyes, no beauty filter. Photorealistic, ultra-detailed, 4K. No text, no logo, no watermark.
Step 2
파리 거리 로케이션
파리 거리 로케이션
2번째로 업로드하면 프롬프트의 @Image 2이 됩니다
생성 프롬프트 · 1845자
A location-scout photograph of a quiet Paris side street in the 6th arrondissement, taken at a 3/4 angle from the middle of the pedestrian pavement looking down the block on a Canon EOS 5D with a 35mm lens. Late morning under a flat overcast sky, twenty minutes after rain has stopped — no sun, no shadows, soft grey light coming from everywhere at once.
Six-storey Haussmann limestone facades run down the left side, cream stone gone grey-streaked with soot at the cornices, tall shuttered windows and continuous wrought-iron balconies on the second and fifth floors, dark grey zinc mansard roofs above. The pavement is wide grey stone slab, generous enough for two people to pass each other with room to spare, edged by a low granite kerb. Beyond the kerb the road is dark wet asphalt with old cobble showing through a patch of worn tarmac; a narrow strip of parked cars sits along the far kerb, well separated from the walking space.
Lived-in detail everywhere: a florist's shopfront halfway down with dark green painted timber and buckets of stems drawn back under the awning, a scuffed grey steel bollard leaning slightly out of true, cigarette ends and a flattened bottle cap pressed into the gutter, rain still beaded on a bicycle chained to a downpipe, a wet plane-tree leaf stuck flat to the stone, standing water in the low points of the pavement throwing soft inverted reflections of the facades.
Muted palette — limestone cream, wet charcoal asphalt, zinc grey, deep bottle green, all desaturated. No warm light sources, no sunbeams, no golden hour, no shop lights on.
Real architecture, physically grounded objects, true light falloff, subtle film grain, slightly imperfect framing like a real scouting snap. Empty of people. NOT a 3D render, NOT CGI, NOT AI-looking. No readable text, no logo, no watermark. Ultra-detailed, 4K.
Step 3
크리스탈 스노우글로브 펜던트
크리스탈 스노우글로브 펜던트
3번째로 업로드하면 프롬프트의 @Image 3이 됩니다
생성 프롬프트 · 2700자
Product reference sheet of one crystal snow-globe pendant necklace as FIVE views on a clean light grey studio background: FRONT view of the globe upright, THREE-QUARTER view, SIDE PROFILE showing the depth of the dome and the full setting, TOP-DOWN view looking into the dome, and a MACRO detail of the bail, the clasp and the chain — plus ONE in-use inset.
The pendant is a flawless rock-crystal dome about three centimetres across, thick and optically clear, held in a polished platinum setting shaped as a fluted pedestal with a fine milgrain edge; a continuous line of tiny pavé-set white diamonds runs around the bezel ring where the crystal meets the metal. The bail above it is a single large round white pearl with a high satin nacre, matching the pearl drops worn at the ears. The chain is a fine platinum cable chain with a small lobster clasp.
Inside the dome, standing on a miniature platinum pedestal, is a single slice of layer cake no bigger than a fingernail — hand-painted enamel in pale butter-yellow and cream, its piped rosettes rendered in raised guilloché enamel, topped with one cabochon ruby cherry on a fine gold stem. The liquid filling the dome is perfectly clear and colourless. Suspended in it, motionless, are flakes of 24-karat gold leaf and slivers of white opal that fracture the light into shifting rainbow spectra — they hang in place rather than settling, as if the liquid had stopped. The miniature reads very slightly too large for its dome, the way objects in a dream do not quite match their scale.
The crystal throws real caustics: prismatic edge refraction along the dome's rim, a small concentrated rainbow cast onto the grey surface beneath, the interior magnified and doubled where the glass curves. Nothing glows by itself — every sparkle is reflection or refraction of the studio light.
Kept, cared for, not new: one soft fingerprint haze low on the crystal, the faintest polish swirl on the platinum pedestal catching the light. In the MACRO view one link of the chain near the clasp is visibly splayed open and fatigued, about to give.
IN-USE INSET — the pendant worn, hanging on its chain against bare collarbones, cropped tight from the base of the throat to the top of the chest so no face is in shot; overcast daylight, the dome throwing a small prismatic scatter onto the skin beneath it.
Shot like a high-jewellery commercial product set on a medium-format camera, soft even light with one controlled specular source, razor-sharp focus, true crystal refraction, true platinum tone, true pearl nacre, true gold-leaf suspension in liquid. Photorealistic, ultra-detailed, 4K. No text, no logo, no watermark, no branding anywhere on the piece.
Step 4
영상 프롬프트
최종본
생성 프롬프트 · 11317자
SCENE CONTEXT
A rain-washed Paris side street under flat overcast light, late morning. A woman walking and talking on a phone is clipped by a passing stranger's elbow, three things leave her in the same instant, time slows almost to nothing so the camera can look at each of them where it hangs, then normal speed returns and everything falls. Controlled multi-shot sequence, seven timed shots, hard cuts only at the times listed below.
ACTIVE REFERENCES
— THE WOMAN. Preserve her face, hair, make-up and her complete costume exactly: the sculpted butter-yellow dress, the sheer lace gloves, the pearl earrings, the ivory satin ribbon crossing up both calves, the cream shoes. Do not use its white studio background, its four-panel layout, its standing lookbook pose or its flat studio lighting.
— THE STREET. Inherit only its architecture, pavement and road materials, wet stone surfaces and overcast light. Do not use its camera position, its framing, or its emptiness — here the street is populated and the camera moves. Do not extend, widen or upgrade the location beyond what the reference shows: no added storefronts, no added ornament, no added vehicles.
— THE PENDANT. Preserve its exact geometry and materials: the crystal dome, the platinum pedestal setting, the pearl bail, the enamel cake miniature with its ruby cherry, the gold leaf and opal slivers suspended in clear liquid, the fine platinum chain. Do not use its grey product background, its multi-view layout or its studio specular lighting. Exactly ONE of this pendant exists in the film.
ADDED and not present in any reference: one paper takeaway cup with one plastic lid, one open silver clamshell phone, one passer-by, one Labrador on one leash.
LOCATION MAP — HOLDS FOR EVERY SHOT
Wide grey stone pavement on screen-LEFT of the frame line, granite kerb and dark wet asphalt road on screen-RIGHT. Limestone facades with wrought-iron balconies run along screen-LEFT behind the pavement. The woman walks toward the camera down the pavement. The passer-by and the Labrador come the opposite way and pass on her right, which is screen-LEFT. After the contact the pendant hangs low-centre near her chest, the cup hangs low on screen-LEFT, the phone hangs high on screen-RIGHT. The camera keeps that arrangement across every cut and never flips it.
FIRST FRAME
Occupied immediately: the camera is already low and already craning, the woman is already mid-stride with the phone already at her right ear and already mid-sentence, the cup already in her left hand, the pendant already swinging at her collarbones. Nothing starts from rest.
FORMAT MODE
Seven timed shots. Real time from 0.0 to 4.4. One single speed change into extreme slow motion at 4.4, holding roughly three percent of normal speed through 12.2, then one single snap back to one hundred percent exactly at 12.2. No ramps anywhere else, no other speed changes, and no slow motion in the final impacts.
OPTICS AND CAMERA
Setup A — handheld-smooth crane on a motion-control arm, starting at ankle height, rising and drifting backward to chest height. Setup B — locked medium-wide on a long lens, static. Setup C — macro dolly on the same arm, moving toward one object and orbiting it through a shallow arc. Setup D — low tilt-down to street level. Lens locked inside each shot; the camera never changes focal length mid-shot.
ACTION TIMING
[00.0-04.4] SHOT 1 — Setup A, one unbroken move, real time. Open low and tight on wet paving stones as she walks through frame: ivory ribbon crossing up her calves, cream shoes striking the stone, water flicking off the leather. The camera cranes up and drifts back with her, passing the ribbon crossings, the round rose-covered skirt swinging against her stride, the two cherries pinned at her waist, the pendant swaying at her collarbones, and arriving on her face. She talks into the phone. At 3.7 the passer-by walking the Labrador comes the other way and that person's elbow clips her upper arm. She recoils backward and three things leave her in the same instant: the chain at her neck gives and the pendant tumbles free on a short outward arc; the cup slips from her left hand and rotates until one edge of its plastic lid lifts away from the rim; the open phone leaves her right hand on a short upward toss. Both hands are empty from that instant, fingers spread, nothing else in either of them.
[04.4-05.4] SHOT 2 — Setup B, slow motion. Medium-wide of the whole accident so the geometry reads. She continues tipping backward a hair at a time, her hair lifting behind her. The pendant, the cup and the phone each creep along their own arc and turn slowly where their own momentum carries them, each staying on its own side of the frame.
[05.4-07.6] SHOT 3 — Setup C on the pendant, slow motion. The camera closes the distance to the pendant and orbits it through a shallow arc: crystal breaking the grey daylight into prismatic edges, gold leaf and opal slivers creeping through the clear liquid inside, the enamel cake slice and its ruby cherry turning with the dome, the snapped chain link trailing behind. She is a soft shape in the background, still going over.
[07.6-09.6] SHOT 4 — Setup C on the cup, slow motion. Macro. Dark coffee escapes only through the real gap between the lifted lid edge and the rim: a thin uneven ribbon of liquid with a few odd-sized droplets trailing it, surface tension pulling each droplet round. A small lateral parallax move around the cup, the lid and the droplets.
[09.6-11.4] SHOT 5 — Setup C on the phone, slow motion. The phone drifts face-down: brushed silver housing, the hinge, half the keypad, both halves open at the same angle they were held at. Its lit screen throws a small blue rectangle onto the wet stone beneath it. The screen itself is never turned toward the camera.
[11.4-12.2] SHOT 6 — Setup B, slow motion. Returns to the exact framing and lens of SHOT 2. She is still going over backward, both hands empty, and the pendant, the cup with its half-lifted lid and drifting coffee, the face-down phone, the one passer-by and the one Labrador are all exactly where they were, none of them nearer the camera than before.
[12.2-15.0] SHOT 7 — Setup D, real time. Gravity takes all three from their own positions. The pendant strikes the wet stone and its crystal scatters light through the splash. The cup hits and bursts, coffee running irregular across the paving. The phone lands face-down and bounces once. The three come to rest well apart from one another. Her back reaches the ground and the camera tilts down and drops with them to street level, ending low on her and the scattered objects among the reflections. Hard cut to black on the last frame.
DIALOGUE AND ACTING
Dialogue language: Japanese. One spoken line only, inside SHOT 1 while her face is frontal and her lips are visible. The woman says into the phone in natural, conversational Japanese, brisk and apologetic, slightly hurried: {はい、もうすぐ着きます。} Her lip movements must precisely match the pronunciation of this Japanese line. Her jaw works, and a small tired smile lands at the end of it. Nobody else speaks and no other mouth moves.
ACTING TASK — she is not performing an accident. Before the contact the work happens in her eyes and her walking rhythm: she is half listening, half navigating the pavement. After the contact she does not act shocked; her body simply goes where it is pushed and her eyes stay open and unfocused.
SCENE DIRECTION (unspoken) — treat this as a small ordinary stumble on an ordinary morning, not as a disaster.
MOTIVE — she is already late and she is telling someone so.
OBJECT LOCK
At 0.0 the only objects in play are: the pendant worn at her collarbones, the paper cup in her left hand, the open phone in her right hand, and the leash in the passer-by's hand. Nothing else is carried by anyone. Only these objects move, and each moves only because a hand releases it or the chain breaks. No object appears, disappears, floats without cause, or morphs into another object. No object is ever carried toward the lens; anything that grows in frame does so only because the camera closed the distance, and no object is redrawn between the wide framing and the macro framing.
PHYSICS
The dress is sculpted fabric imitating piped cream, never edible cream: it creases and springs back like stiff couture and never smears, melts, or transfers onto anything it touches, including during the impact and the fall. Walking: heel lands first, strict left-right alternation, one foot always on the ground until the recoil. The pendant has real glass mass, dropping in a tight arc and ringing hard on stone. The paper cup is light and crushes on impact while the plastic lid stays attached along one edge. The phone is light plastic and metal, landing flat with one small bounce. The coffee is a real liquid with real viscosity: it does not explode outward, it leaks and then runs.
LIGHTING
Flat overcast daylight for the whole sequence. No lamp, sign, screen or headlight turns on or off at any point, and no light source changes at the speed change. Every brightness change in frame is occlusion, never a lighting event. Dominant colour is desaturated limestone grey and wet charcoal; the only saturated colours in frame are the butter yellow of the dress, the red of the cherries, and the rainbow that the crystal breaks out of the grey light.
AUDIO
No background music at any point. Diegetic sound only. Heels on wet stone, ribbon and stiff fabric brushing, liquid shifting inside the cup, one car passing far off, dog claws scratching stone, the dull knock of the elbow, one thin metallic snap. From 4.4 the ambience drops away to near silence under a single low sustained tone, and at 12.2 everything returns at once: crystal ringing on stone, the cup bursting, the flat knock of the phone, one bark. Spoken audio is the one quoted Japanese line only, delivered frontally with her lips visible.
STYLE AND QUALITY
Ultra-photorealistic live action, premium fashion film. Sharp clarity, stable picture, natural colour, readable face. Real skin with visible pores and natural catchlights, real hair with distinct roots, real fabric weight, real glass refraction, real liquid behaviour, real inertia. Shallow photographic macro depth of field in the close shots, with the crystal, the droplets and her face staying razor-sharp through the slow passage. Subtle film grain, controlled highlights, restrained editorial grade. No ghosting, no duplicated limbs, no identity drift across cuts. Never render a reference sheet or duplicate a subject.
POSITIVE CONSTRAINTS
1. Exactly one woman, one passer-by, one Labrador, one leash, one pendant, one cup, one lid, one open clamshell phone. Everyone else on the street is a distant unnamed extra whose face resembles no reference.
2. Her identity from holds through the contact, the recoil and the fall, especially in the macro shots where she is out of focus behind the objects.
3. Both hands stay empty from 4.4 to the end, and nothing replaces either prop.
4. The pendant, the cup and the phone stay on their own sides of the frame and land in three separate places.
5. The phone screen is never turned toward the camera in any shot.
6. No subtitles, no on-screen text, no logo, no watermark throughout.