previs-control-passes
previs-control-passes
navarisun1982/comfyui-motion-previs-studio (MIT) is a local WebGL/Electron app. The user installs and runs it from its README. This skill does not install or run it. Its architecture and template docs are copied in references/ (MIT).
Flow
- Block the shot in the studio: actors, motion presets, camera, lens, aspect.
- Use Previs Render & Export to export a 24 fps composite MP4, a clean beauty pass, an OpenPose 18-joint MP4, a normalized depth MP4, a subject alpha mask, and a bundle with the
4×4camera matrices and the compiled prompt. - Route the passes:
- OpenPose or depth go to
control-to-video(Wan VACE pose, LTX-2 depth or pose). - Camera matrices go to
camera-control(ATI or Fun Camera) or are kept for continuity. - The alpha mask goes to
background-matteorvideo-inpaint-remove.
- OpenPose or depth go to
- Record the pass files on the shot in the video-director manifest (
shot.extra[].params).
Template bindings (scripts/binding_apply.py)
The studio pairs each API graph with a <id>.binding.json that maps fields (prompt, seed, frames and so on) to {node, input} and verifies them before queueing. The studio's own binding files live in its repo under workflows/templates/. scripts/binding_apply.py (stdlib, original) implements the same contract for any graph:
python scripts/binding_apply.py check <graph.api.json> <binding.json> checks that each node exists, the class_type matches and the input exists. Then python scripts/binding_apply.py apply <graph> <binding> prompt="..." seed=5 --out filled.json. Fields marked fixed are always forced to their value. Fields with a studio-specific patch (for example timeline) are refused, so edit those inputs by hand.