sirioberati/Genjustsu-Open-Source-Workflow

Loading README…

Layer	Technology and purpose
Local app	Python 3.12+, FastAPI, Uvicorn; HTML, CSS and vanilla JavaScript UI
Video/audio processing	PyAV, NumPy, Pillow, OpenCV and SciPy; local encoding, compositing and audio restoration
Vocal isolation	Hosted Demucs / htdemucs through Replicate
Pitch	Rubber Band CLI; isolated vocals +3 semitones without changing duration
Colored depth	Replicate `lucataco/depth-anything-video`; optional Hugging Face Video Depth Anything or `chenxwh/depth-any-video`
Segmentation	Replicate `lucataco/sam3-video`; raw per-frame masks and mandatory visual review
Face mesh	Local Google MediaPipe Face Landmarker; saved landmarks and overlay, isolated Python 3.12 worker environment
Generation	Enhancor Seedance; Edit for image references, Omni for video references; draft first, approved 1080p upgrade
Media hosting	Tmpfiles primary; configurable Catbox fallback; direct-link checks before generation
Completion	Persisted job files and polling; callback-only Cloudflare Quick Tunnel for the provider callback requirement
Agent workflow	Included AGENTS.md, setup runbook, skill and CLI; a user-supplied coding agent performs visual QA

Genjustsu — Open Source Workflow