시작은 해외에서 돌던 원테이크 프롬프트 하나였습니다. 뉴욕 한복판을 달리다 넘어지는 순간 시간이 멈추는 구조인데, 그대로 쓰지 않고 무대를 대낮 전통시장으로 통째로 옮겼습니다. 정지 화면의 밀도 때문입니다. 어묵 솥에서 오르다 만 김, 막 터진 뻥튀기 구름, 전 부치던 기름방울까지, 시장은 멈추는 순간 공중에 떠 있는 것이 많은 장소입니다.
영상은 처음부터 끝까지 한 번도 끊기지 않습니다. 앞 8초는 2000년대 초 가정용 캠코더로 찍은 척하는 추격 화면입니다. 촬영자가 같이 뛰어서 화면이 계속 흔들리고, 사장님 독촉 전화와 숨소리가 그대로 들어갑니다. 넘어지는 순간 세상이 얼어붙으면 같은 카메라가 손떨림의 관성을 그대로 이어받아 떠오르고, 1인칭 시점(FPV) 드론처럼 정지된 세계를 누빈 뒤 처음 구도로 돌아와 시간이 다시 흐릅니다.
잠근 것은 레퍼런스 세 장입니다. 정장에 은테 안경을 쓴 주인공 캐릭터 시트, 통화 화면까지 그려 넣은 2000년대 휴대폰 소품 시트, 그리고 서류 소품입니다. 서류는 원래 만들 생각이 없었습니다. 원본 프롬프트처럼 글자를 영상 모델이 직접 그리게 했는데, 한글이 버티지 못했습니다. '댓글'이 '덷금'으로, '프롬프트'가 '프럼프트'로 깨져 나왔습니다. 그래서 명조체 문장을 인쇄한 서류를 이미지로 먼저 만들어 물렸습니다. 소품 글자도 한 번 데였습니다. 화려한 장식 위에 글자를 얹으라고 했더니 스티커처럼 떠 보여서, 잉크가 종이 섬유에 스며든 질감을 지시에 박고서야 인쇄물로 읽혔습니다.
각 단계의 프롬프트를 복사해 이미지를 만들고, 이미지마다 붙은 @Image 번호 순서 그대로 업로드하면 마지막 영상 프롬프트의 번호가 정확히 맞습니다.
Step 1
인물시트
이 단계에서 할 일
프롬프트를 복사해 이미지 생성 도구에 붙여 넣으세요
「내 것으로 바꾸기」가 붙은 이미지는 내 사진이나 내 제품을 함께 올려 만드세요
만든 이미지는 @Image 번호대로 저장해 두세요 — 영상 단계에서 그 순서로 올립니다
인물시트
이 자리에 내 사진이나 내 제품을 넣어 만드세요
1번째로 업로드하면 프롬프트의 @Image 1이 됩니다
생성 프롬프트 · 1739자
Character reference sheet as THREE panels side by side, shot in a sunlit Korean traditional-market side alley at midday — bright natural daylight bouncing off vinyl awnings and worn pavement, soft open shade. PANEL 1 — full-body FRONT view of a woman standing naturally, a statuesque full-height figure captured head to toe, long legs taking up well over half of her standing height, never compressed or shrunken to fit the panel. She wears the complete outfit: a sharply tailored charcoal-grey double-breasted pantsuit with structured shoulders and peaked lapels, a crisp white dress shirt, slim straight-cut suit trousers with a pressed crease, and polished black leather loafers with a low block heel. Worn accessories: thin silver rectangular eyeglasses, a slim steel wristwatch, small silver stud earrings; dark hair in a neat low bun with a few loose strands. PANEL 2 — full-body BACK view, same outfit, same spot, showing the double-vented blazer back, the low bun, and the natural trouser drape. PANEL 3 — large head-and-shoulders face close-up, identity strictly from the attached reference face, wearing the thin silver rectangular eyeglasses, alert focused expression, real skin texture with visible pores and natural catchlights in the eyes. The attached photo provides facial identity only — do not carry over its camera angle, framing, pose, expression or lighting; every panel follows the standing production-still staging described here. Same person, same outfit, same lighting in all three panels. Her hands are empty in every panel. Looks like real production stills from a drama set — real fabric weight and wrinkles, no beauty filter, no CGI. Photorealistic, ultra-detailed, 4K. No readable text, no logo, no watermark.
Step 2
효과 · 이미지 2장
이 단계에서 할 일
프롬프트를 복사해 이미지 생성 도구에 붙여 넣으세요
만든 이미지는 @Image 번호대로 저장해 두세요 — 영상 단계에서 그 순서로 올립니다
레트로폰
2번째로 업로드하면 프롬프트의 @Image 2이 됩니다
생성 프롬프트 · 1055자
Product reference sheet of a generic early-2000s candybar mobile phone as THREE views on a clean light-grey studio background: front view with the small low-resolution color LCD lit up, a 3/4 side view showing the chunky keypad and body thickness, and a back view showing the plain matte casing and speaker grille — plus ONE in-use inset of a hand gripping the phone pressed against an ear mid-stride. The lit screen shows authentic early-2000s mobile phone UI graphics: a small green call icon, call duration "2:38", and the contact name "사장님" in a chunky pixelated Korean font. All on-screen text must be in Korean only — no text in any other language. Generic no-brand design: silver-grey plastic body, slightly worn keypad with faded number paint, no logos or brand markings anywhere on the device. Exact geometry, true materials, fine surface wear from daily handling. Shot like a commercial product photo set, soft even light, razor-sharp focus. Photorealistic, ultra-detailed, 4K. No logo, no watermark, no text anywhere except the phone screen UI.
CTA서류
3번째로 업로드하면 프롬프트의 @Image 3이 됩니다
생성 프롬프트 · 925자
A single sheet of plain white A4 office paper photographed straight-on, top-down, filling the frame, lying flat on a neutral desk surface — an ordinary printed office document. Centered on the page, one bold sentence printed in a classic Korean serif (Myeongjo-style) typeface, deep black ink, large enough to dominate the page across two lines: 댓글에 'AI' 남겨주시면 프롬프트 보내드릴게요. The lettering is real laser-printed ink absorbed into the paper fiber — crisp edges with subtle toner texture, the paper grain visible through the strokes, never a floating digital overlay. Every Korean character must be accurately and correctly formed, perfectly legible. All text in Korean only — no other words anywhere on the page. Slight natural paper curl at one corner, true paper texture, soft even daylight, razor-sharp focus on the print. Photorealistic, ultra-detailed, 4K. No logo, no watermark, no text except the single printed sentence.
Step 3
편집 마무리
최종본
생성 프롬프트 · 6276자
defines the protagonist: her face, low-bun hairstyle, thin silver rectangular eyeglasses, and complete outfit — charcoal-grey double-breasted pantsuit, white shirt, black leather loafers. Do not copy the sheet's standing poses, panel layout, or background framing.
defines the exact silver early-2000s candybar mobile phone, including its lit LCD call screen: green call icon, call duration "2:38", contact name "사장님". Do not use its grey studio background or panel layout.
defines the printed document: its exact typography and the printed Korean sentence on white A4 paper. Do not use its desk background.
ONE CONTINUOUS SHOT — no cuts, no transitions, no hidden edits, from first frame to last.
[Global Settings]
A crowded Korean traditional market alley at bright midday: a narrow lane packed with shoppers and vendors, steaming fish-cake vats, oil sizzling on griddles, stacked produce crates, hanging goods, a delivery scooter idling, sunlight cutting between vinyl awnings. The whole video looks like authentic early-2000s consumer camcorder tape: small-sensor softness, limited dynamic range, slightly muddy colors, mild highlight smear, imperfect white balance, tape compression, no stabilization, no modern digital sharpness, no cinematic color grade. Every object in frame is period-correct for the early 2000s — stalls, carts, scooter, clothing, every device pre-2005. All background signs and price boards stay soft and illegible.
From the first frame the protagonist visibly carries everything: a matte black leather briefcase clamped under her left arm, a clear plastic cup of iced americano plus a few loose white printed documents — the front one being the document — gripped in her left hand, and the phone pressed to her right ear with her right hand. No object appears, disappears, floats, or morphs except through her fall.
[Stage 1 | 0-8s — handheld chase]
An unseen cameraman physically runs after her, filming from her left side, framing roughly waist-up: heavy natural handheld shake from running, irregular footsteps, audible breathing, sudden reactive reframes, shoppers occasionally passing between camera and subject for a fraction of a second. She runs urgently through the crowd, turning her shoulders to squeeze through gaps, overloaded, nervous, breathless.
Dialogue language: Korean, natural conversational delivery.
0-2s — the boss's voice through the phone's tiny speaker, thin, compressed, authoritative: {프롬프트 구했어?}
2-5s — she answers into the phone, breathless and urgent: {지금 받으러 가는 중이에요!}
5-7s — the boss, low and threatening through the speaker: {한 시간 안에 없으면… 알지?} — during this line the cameraman drifts slightly ahead, catching more of her face.
7-8s — she rushes to reassure him: {네네 사장님, 벌써 거의—} — she never finishes the sentence.
Audio throughout: genuine location sound from a consumer camcorder microphone — vendor calls, sizzling oil, bubbling broth, footsteps, her breathing, the thin phone-speaker voice.
[Stage 2 | at 8s — the trip, then time freezes]
At 8 seconds her foot catches on a warped floor mat between stalls and she pitches forward: the cup lid pops off, ice cubes and an arc of coffee launch outward, the documents fan into the air, her glasses slide off her face, the phone leaves her ear, the briefcase swings away from her body. At the exact instant before she hits the ground, TIME STOPS COMPLETELY. Every person in the market freezes mid-stride. The steam column above the fish-cake vat freezes solid in the air. A puffed-rice cannon at a nearby stall has just burst — its white cloud of puffed grains hangs suspended mid-explosion. Oil droplets hover above a griddle. A pigeon is frozen mid-flap. She is suspended mid-fall among frozen ice cubes, a flowing coffee arc, hovering papers, her floating glasses, the floating phone, the swinging briefcase. All ambient sound cuts dead to total silence, leaving only a faint low hum.
[Stage 3 | 8-17s — seamless FPV through frozen time]
The same camera, with no cut and no visual transition, carries its handheld momentum upward and organically becomes a fast, precise FPV drone movement — no visible drone or operator, wide-angle perspective, deep focus, physically believable inertia.
8.5-10s: rise above her, revealing the frozen market wide from above.
10-11.5s: dive to her face — frozen expression of pure panic, the silver rectangular glasses suspended just off her nose. Hold briefly.
11.5-13s: dart through the frozen coffee arc — individual ice cubes and droplets hanging motionless, the cup lid floating nearby, daylight refracting through them.
13-14.5s: lock onto the floating phone's small lit screen, matching exactly and clearly readable: green call icon, "2:38", contact name "사장님".
14.5-16.5s: glide between the suspended sheets of paper to the front-most document, which matches exactly — the same typography and the same printed Korean sentence, perfectly legible and filling the frame: 댓글에 'AI' 남겨주시면 프롬프트 보내드릴게요.
16.5-17s: pull away in a smooth descending arc around her frozen fall and settle back into the exact side-on medium framing of the opening handheld shot.
[Stage 4 | 17-20s — time resumes]
The instant the camera settles, time resumes naturally with no flash, no dissolve, no effect: the full market noise crashes back all at once. She hits the ground — coffee splashes across the pavement, ice cubes scatter, papers rain down, the glasses and phone clatter, the briefcase thumps. Natural handheld shake returns as the now-standing cameraman keeps filming her sprawled on the pavement; two nearby vendors turn toward her in surprise.
[Maintain Consistency]
She remains the same single person as throughout — same face, glasses, hair, suit. Crowd count and vendor positions stay consistent; every frozen person and object stays perfectly motionless from 8s to 17s and resumes from its exact frozen position. No object appears out of thin air, disappears, duplicates, or morphs. No subtitles, no floating captions or graphics, no logo, no watermark — the only readable text in the entire video is the phone screen UI and the document sentence, both in Korean; all other text stays illegible. No background music; diegetic location sound only.