openjutsu
영상에서 원클릭으로 인물 교체, ComfyUI·MiniMax H3 기반 오픈소스 로컬 힉스필드 겐주츠 대안, FLUX.3 Klein 재조명 지원
저장소 정보는 아직 수집하지 않았습니다.
아직 감지된 업데이트가 없습니다. GitHub 릴리스가 나오면 여기에 보입니다.
Turn a list of shots into a finished film, without wasting money on bad takes
저장소 README에서 가져옴
README 전체를 불러오는 중입니다.
README 전문 보기 ↗Turn a list of shots into a finished film, without wasting money on bad takes
Describe each shot in JSON. vs makes the AI clips, then cuts them into one video.
Setup, costs, commands, and a sample film you can run.
You need Node 24+ and ffmpeg (brew install ffmpeg).
git clone https://github.com/mblode/video-studio.git
cd video-studio
npm install
npm link
Copy .env.example to .env, add your AI_GATEWAY_API_KEY from Vercel AI Gateway (default Seedance 2.5) or ARK_API_KEY from the BytePlus console (Seedance 2.0 / lighthouse), then run vs doctor.
Try the sample film in films/lighthouse/:
vs cast sync films/lighthouse/shots.json
vs generate films/lighthouse/shots.json --dry-run
vs stills films/lighthouse/stills.json
vs animatic films/lighthouse/shots.json
vs generate films/lighthouse/shots.json --max-cost 18
vs stitch films/lighthouse/shots.json --xfade 0.4
Start your own with vs init films/my-film.
Describe each character once in characters.json, list them in a shot's
cast, and vs cast sync writes the rest: a multi-view character sheet into
stills.json, the reference into the shot, and the binding sentence that tells
the model which @Image is whose face. It counts the ordinals for you, which
is the part that is easy to get wrong and expensive to discover.
Sync writes literal text into shots.json and stills.json — nothing at
generate time reads characters.json — so you review the diff before you spend,
and vs cast sync --check fails a build that has drifted.
--max-cost stops a run that would go over.vs use picks which one to use.npx skills add mblode/video-studio -g --agent codex claude-code -y
Works with Claude Code, Codex, Cursor, and other skills.sh agents. From a clone, they are already available.
MIT
Crafted by Matthew Blode