늦잠에서 깬 여자가 식빵을 물고 창밖으로 날아오릅니다. 목적지는 회사입니다.
30초를 한 번에 갔습니다. 이야기를 굴리는 장치는 하나입니다. 필사적으로 날아온 끝이 하찮을 것. 빨래 홑이불에 얼굴을 박고, 식빵을 노린 까마귀와 빌딩 사이에서 공중전을 벌이고, 고가 전철 창 높이에 붙어 승객들에게 어정쩡하게 인사하고, 배달 드론과 말없이 경주합니다. 구름을 뚫고 올라가 아침 햇빛 위를 활공한 뒤 역 광장에 삼점 착지까지 성공하는데, 출근 도장을 찍으려고 꺼낸 휴대폰을 보고 표정이 굳습니다. 「…어라, 오늘 쉬는 날?」
고정한 것은 레퍼런스 이미지 한 장입니다. 앞과 옆과 뒤를 한 줄에 세운 캐릭터 시트에 얼굴 확대 칸을 붙인 것으로, 얼굴과 머리와 착장을 전부 여기서 가져옵니다. 대신 시트의 흰 스튜디오 배경과 중립 포즈는 화면에 쓰지 않는다고 못을 박았고, 착장을 언제부터 입는지는 레퍼런스 선언이 아니라 따로 세운 시간표에만 걸었습니다. 그래야 잠옷으로 시작하는 첫 2초가 살아남습니다.
비행에는 보이는 장치를 하나도 두지 않았습니다. 대신 결과만 남겼습니다. 이륙마다 웅크림과 반동이 있고 그 바람에 화분이 넘어지고 빨래가 펄럭입니다. 바람은 스카프와 치마와 머리를 계속 끌고, 착지는 무릎이 충격을 받습니다. 빛나는 기운이나 망토 같은 것은 한 번도 쓰지 않았습니다.
속도를 바꾸는 자리는 두 군데로 못 박았습니다. 거울 앞 환복은 방과 거울을 완전히 고정해 두고 사람만 빠르게 움직입니다. 구름 위에서는 반대로 사람을 정상 속도로 고정하고 세상만 빠르게 흘려보냅니다. 두 곳 다 서서히 풀지 않고 툭 끊어 원래 속도로 돌아옵니다.
단계 2개 · 레퍼런스 1장
따라 만들기
각 단계의 프롬프트를 복사해 이미지를 만들고, 이미지마다 붙은 @Image 번호 순서 그대로 업로드하면 마지막 영상 프롬프트의 번호가 정확히 맞습니다.
Step 1
갸루 4각 캐릭터 시트
갸루 4각 캐릭터 시트
1번째로 업로드하면 프롬프트의 @Image 1이 됩니다
생성 프롬프트 · 2228자
Character reference sheet as FOUR full-body views in a row on a clean white studio background — FRONT, 3/4, SIDE PROFILE, BACK — plus a large face close-up inset panel, front-facing. Subject: a woman with voluminous dark-brown spiral-curl hair falling past the shoulders, wearing a sheer pale-blue mesh off-the-shoulder long-sleeve top with a subtle tonal sheen and thumbhole cuffs, layered with a matching pale-blue draped shawl whose long soft tails hang from the shoulders down past the hips, a white tiered mini skirt with dense ruffled lace hems and a white lace-trimmed under-short peeking below the lowest ruffle, a silver chain belt slung on the skirt waist with a large rhinestone star buckle at the front hip and a few tiny dangling charms including a small blue cross charm, white floral-lace tights, white leg warmers with scattered black dalmatian spots slouched over the ankles and tied with light-blue ribbon bows, and white platform round-toe strap shoes with black soles and straps. Worn accessories: a silver necklace with a star pendant and a long fine drop chain hanging to the sternum, a stacked mix of beaded and pearl bracelets on one wrist and thin silver bracelets on the other, and a small black-and-white spotted puppy plush charm clipped to the chain belt at the hip. Identity strictly from the attached reference face — long defined curled lashes, flushed rosy cheeks, glossy pink lips. The attached street snap provides facial identity only (features, skin, hair colour and curl texture); its high-angle camera, cropped framing, tilted pose and outdoor asphalt lighting must not carry over — all four views and the inset follow the neutral standing lookbook framing described here. Tall runway-model proportions of about nine heads, small head and long legs taking up well over half of each figure, every view spanning the panel from the top of the hair to the shoe soles with no squashing or shrinking to fit. Same person, same outfit, same neutral standing pose logic in every view. Clean even studio lighting like a fashion lookbook turnaround, true fabric colour, real mesh transparency, honest ruffle volume and knit texture. Photorealistic, ultra-detailed, 4K. No text, no logo, no watermark.
Step 2
영상 프롬프트
영상 프롬프트
생성 프롬프트 · 7157자
is the ONLY reference for the main character. It defines her identity — preserve her exact face, makeup, and voluminous dark-brown spiral-curl hair in every shot — and it defines the design of her outfit: sheer light-blue off-shoulder top, long light-blue scarf, silver star necklace, white lace ruffled miniskirt, white lace tights, dalmatian-print leg warmers with blue bows, chunky black-and-white platform shoes. WHEN she wears this outfit is set only by the wardrobe timeline below. Do not use the sheet's white studio background, panel layout, or neutral standing poses. The same single woman appears throughout — no identity drift, no second copy of her.
WARDROBE TIMELINE: 0–2 sec — rumpled plain pajamas, curls extra messy from sleep, no accessories. 2–6 sec — she dresses during the mirror timelapse described below. From 6 sec to the end — the exact complete outfit from , unchanged.
Create a chaotic 30-second Japanese live-action comedy video. Fast and physical, no dead time, rapid cuts and dynamic camera movement. All dialogue is Japanese only, and only she speaks — no other on-screen mouth ever syncs to any voice. Her voice is cute, high-pitched and bright.
STYLE: Warm faded early-digital handheld footage with low-saturation color, soft focus, mild noise, slightly blown highlights and candid movement. No timestamp, date stamp or UI overlays. Not modern crisp digital. Speed changes exist ONLY in the two designated timelapse segments (2–6 sec and 20–23 sec); no slow motion anywhere, no speed ramps anywhere else.
FLIGHT RULES: Her flight is played as physical and live-action. Every takeoff has a visible crouch, recoil and downdraft that shoves nearby objects; wind constantly drags her scarf, skirt and curls; landings carry real impact through her legs. She flies clumsily, like a new driver. No glowing aura, no energy effects, no cape, no superhero costume, no named superhero reference.
SOUND: Playful upbeat Japanese gyaru dance-pop begins at the first frame and continues throughout. Sync cuts, the alarm clock, wardrobe snaps, the takeoff downdraft, whipping wind, flapping laundry, crow caws, train rumble, drone buzz, rain patter, landing impact and her footsteps to the beat. No silent sections.
SETTING: A small messy Japanese bedroom with a bed, an alarm clock, a full-length mirror and clothes piled around; a slice of toast on a plate by the door. Outside: a Japanese residential street with laundry lines hung with white sheets, power lines, then a downtown district of office buildings, an elevated commuter train line, one small delivery drone, one crow, one small grey rain cloud, and a station square with ordinary Japanese pedestrians whose faces and outfits all differ. Japanese signage stays in the background.
00:00–00:02 — WAKE-UP
Extreme close-up of her eye snapping open under messy curls. She bolts upright in bed, grabs the alarm clock, stares at it and gasps. She kicks the blanket off.
00:02–00:06 — DRESSING TIMELAPSE
Locked-off camera facing the full-length mirror. The room and the mirror frame stay pin-sharp and completely motionless — nothing in the background moves. Only she moves, in fast-forward with slight motion trails: pajamas off, top on, skirt on, scarf thrown, leg warmers yanked up, shoes buckled, necklace on, bag slung across. At 5.5 seconds the speed snaps back to normal with no ramp: she stands fully dressed in the exact outfit, jams the slice of toast in her mouth and strikes a quick check pose.
00:06–00:08 — TAKEOFF
On the front step she crouches deep, then launches vertically like a rocket, toast still in her mouth, mumbling through it: 「いってきま〜す！」 The downdraft knocks over a potted plant and sets the laundry flapping.
00:08–00:11 — LOW-ALTITUDE OBSTACLES
Shot like FPV drone chase footage from here through the sky sequence. She weaves clumsily between laundry lines, flies face-first into a hanging white sheet that wraps over her head and over the lens — a beat of white fabric filling the frame — then the sheet peels away and she is suddenly weaving through a downtown canyon of office buildings, dodging under a power line with a twisting roll.
00:11–00:14 — CROW DOGFIGHT
One crow locks onto the toast in her mouth and dives at her. A tight aerial chase: she banks hard, corkscrews between two buildings, the crow snapping at the toast. She swats it away, takes a defiant big bite of the toast, and the crow peels off.
00:14–00:17 — TRAIN FORMATION
She levels out beside an elevated commuter train, flying parallel at window height. Passengers inside raise their phones toward the glass; she gives an awkward mid-air bow. The train slides into a tunnel and she is left bowing to empty air, catches herself, and speeds up embarrassed.
00:17–00:20 — DRONE RACE
A small delivery drone pulls up alongside her. They glance at each other. A silent race breaks out — she pumps her arms, the drone buzzes harder — then she blasts forward past it and looks back smug.
00:20–00:23 — CLOUD-TOP GLIDE TIMELAPSE
She punches up through a small grey rain cloud — a splash of rain soaks her for half a second, frame briefly whiting out inside the cloud — and breaks into golden morning sunlight above a sea of clouds. She holds a steady spread-arm glide pose at fully normal speed, hair drying in the wind: she is the rigid anchor. Only the world moves in timelapse — the cloud sea and the city far below stream past at high speed, cloud shadows racing across rooftops. At 23 seconds the timelapse snaps off with no ramp as she tips into a dive.
00:23–00:26 — PERFECT LANDING
She dives toward the station square and nails a clean three-point landing, skirt settling, impact flexing through her knees. She rises, sweeps her curls back, checks her wristwatch and does a tiny victory step — made it on time.
00:26–00:30 — PUNCHLINE
She pulls out her phone to clock in. Freeze. Her face goes perfectly deadpan. She says flatly: 「…あれ、今日休み？」 One visible sigh. She crouches and launches vertically straight back the way she came, downdraft scattering a few leaves. The camera holds on the empty spot in the square for one beat. HARD CUT TO BLACK on the beat.
ACTING KEYWORDS: jolt awake, alarm shock, fast-forward dressing, toast in mouth, crouch launch, clumsy weaving, sheet over face, twisting dodge, aerial dogfight, defiant toast bite, awkward mid-air bow, embarrassed speed-up, silent drone race, smug look back, soaked in cloud, sunlit glide, hair drying in wind, three-point landing, victory step, deadpan realization, flat sigh, vertical relaunch.
NEGATIVE: no identity drift, no face change, no second copy of the main character, no outfit change after 6 seconds, no glowing aura, no energy effects, no cape, no superhero costume, no named superhero, no cartoon effects, no visible wires, no crow touching her face, no injury, no blood, no speed changes outside the two designated timelapse segments, no slow motion, no static shots, no dead time, no mature voice, no deep voice, no subtitles, no on-screen text, no logo, no watermark, no timestamp, no UI, no modern crisp digital look, no CGI look, no fade before the final frame.