No indoor smoking.
45.25 s
Production
- Images
- ChatGPT
- Video
- Google Flow · Omni 1.1 Flash / Veo 3.1 Lite
- Sound effects
- ElevenLabs
- Music
- Suno V6
- Editing
- FFmpeg
- Story reference
- @mohammedkhattab55
Scenes
One frame.
카메라 앞에서 오른손과 슬리퍼가 급진입한다. 천 윗면은 렌즈 쪽, 고무 바닥은 벽 쪽이며, 손은 뒤꿈치의 잡는 면을 가린다.
Settings
- Model
- Omni 1.1 Flash
- Requested duration
- 4 s
- Source
- 360p
- Ratio
- 9:16
- Control
- first-and-last-frame
The prompt alone did not deliver the final impact speed. Compress the first 5 source frames to 1 output frame (5×), then use 4 contact frames and 42 live hold frames. Add one-frame 18% flash, four-frame damped shake (14, −9, 5, −2 px), six sharp lens specks and mild background blur. Keep the small agi.geek title over the hand; do not black out.
Input images
First-frame input
Last-frame inputVideo prompt
STRIKE v23 RIGHT HAND ANATOMY CORRECTION. Omni 1.1 Flash 360p 4초 9:16 단1개 4크레딧. FIRST v23-right-front-start.png, LAST v23-right-contact-end.png. Use ONLY these two attached images, no prior videos. Create ONE4-second vertical9:16video. At0.00s exactly match FIRST: huge motion-blurred foreground slipper already attacking. Anatomically RIGHT hand throughout: forearm enters BOTTOM RIGHT, back of right hand toward camera, THUMB wraps SCREEN LEFT edge, four curled finger knuckles SCREEN RIGHT. Same grip in LAST image. Do not mirror whole scene or regenerate a left hand. The right palm covers rear/heel opening/gripping face, not a rim pinch. Fabric UPPER faces camera; RUBBER BOTTOM faces wall away from lens. The hand and slipper stay firmly joined, no autonomous slipper. Maintain same charcoal CHECK woven cloth, TWO transverse bands, flat rounded toe, thin dark sole. FRONT-TO-BACK strike along depth axis AWAY from camera into wall. Apparent size quickly DECREASES as huge foreground reaches wall. No sideways sweep, camera zoom or camera motion. Fixed amber fibrous wallpaper/light. Background insect/can/cigarette is occluded by strike, no flight/growth. At0.00s already thrusting, no idle preparation. From0.00s to0.0833s drive immediately away from lens to wall. At0.0833s rubber BOTTOM makes complete wall contact, SAME RIGHT HAND/thumb-left/forearm-right holding heel. From0.0833s to0.2083s a sharp stop and minimal material compression. From0.2083s to4.00s remain in LAST contact pose with tiny live hand tension, never lift/reveal/strike again. No slow pushing or easing, no hand switching/morphing, no gloves or extra hand. Light stays ON; no text/blood/speech/music. A sharp dry smack at0.0833s, then quiet.
Google Flow & sound
한국어 제작 요약: 동일한 방 마스터와 인물 B를 먼저 고정한다. 맨발·민소매·체크 반바지·슬리퍼 두 짝의 위치를 컷마다 대조하고, 실제 입력 첫/끝 이미지를 아래에서 확인한다.
In Google Flow, use the exact first/last image controls recorded per shot. The human world uses an ordinary iPhone-like home-video look; the mosquito world uses gritty amber macro materials. These are separate shots, not an insect growing inside a human shot.
Omni 1.1 Flash takes started at 360p. Veo 3.1 Lite takes used native 720p. Several selected 360p takes were later upscaled in Flow. The final 720 × 1280 export does not make every source native 720p.
Do not assume the requested duration equals the downloaded duration. In particular, the slipper pickup requested 4 seconds but returned 8. Inspect the file, then use frame-based in/out points.
Suno V6 music prompt used in production: Intimate sad instrumental film cue, sparse felt piano and a very soft low cello, 62 BPM, simple minor-key melody, exhausted lonely late-night feeling, sincere and vulnerable, warm slightly imperfect room sound, immediate recognizable melody from the first second, restrained dynamics, no long intro, no big orchestral build, no percussion, no drums, no beat drop, absolutely no vocals, no humming, no spoken words. Short 45-second cue suitable under quiet crying and a deadpan dark comedy; gentle unresolved repeating motif, leave space for sound effects.
62 BPM and 45 seconds were prompt requests, not measurements. The selected Suno result was approximately 69.96 seconds. The final film uses selected segments at 14.25–21.75, 24.50–26.25, 28.7917–32.0417 and 36.0417–37.625 seconds; human shots have no sad-music stem.
ElevenLabs supplied the mosquito buzz, switch and impact effects. Can and sip sounds also reuse generated scene audio. The impact is louder than the other effects; keep the final master below clipping. No audio or video download is offered here.
Editing
최종 타임라인은 24 fps / 1086 frames / 45.25 s. 집기 장면은 1초로 줄이고, 캔 개봉은 다른 손이 보이는 프레임을 덜어 시간 생략으로 연결했다. 건배는 벽에 닿기 전에 끊는다. 모기 컷에서만 슬픈 음악을 이어 붙이고, 마지막 4프레임은 모든 소리를 끊은 뒤 1프레임 급타격과 큰 짝 소리를 배치한다. 마지막 손 화면 위에 작은 agi.geek·블러·렌즈 피 방울을 합성했다. 이어 국토교통부·LH 생활수칙 포스터의 선택한 상단만 1초간 두 바퀴 회전·확대하고 4.5초 정면 유지한다. 기관 표기는 보존한다. No closing spoken line was recorded; the final film contains no added narration. This guide preserves the original generation prompts and explains subsequent edits separately. Final colour: user-selected restrained B grade on human shots only. Preserve the mosquito amber materials, the unlit opening and the original audio. Human grade: colorbalance rs=-0.008, bs=0.008, rm=-0.008, bm=0.006 with preserved lightness; contrast 0.985, saturation 0.92, brightness +0.002.
By agi.geek · Updated 2026-10-19.
Prompts are the originals used in production.
Workshops & production enquiries · All works