StudioCast open_video model pack templates

This directory contains metadata-only templates for open-source video model packs.

StudioCast does not commit model binaries into git.

Installed model packs live at:

• ${XDG_DATA_HOME:-$HOME/.local/share}/studiocast/models/open_video/

For sourcing, conversion, and hosting of the actual model binaries (ONNX / dlib data files), see:

• docs/open_video_model_conversion.md

Subject areas

• segmentation/ — matte/alpha models (task = "matting")
• face_detection/ — face detection models
  • Templates include YuNet in Good/Better/Best variants (int8 / int8bq / fp32).
• face_landmarks/ — facial landmark detectors
• video_denoise/ — video denoising models
  • Templates include FastDVDnet sigma presets in Good/Better/Best variants (sigma15 / sigma25 / sigma50).
• eye_contact/ — gaze/eye contact models
  • Templates include gaze-correction-cam in Good/Better/Best variants (FP16 / FP32 / v3 placeholder).

What’s in a pack?

A pack is a directory containing:

• model.json — pack metadata (see docs/open_video_model_packs.md)
• LICENSE.txt — upstream licensing summary
• model artifacts (downloaded at install time), e.g.
  • *.onnx
  • shape_predictor_68_face_landmarks.dat