LP-Distilled0.1-ONNX by slperez
Model: https://huggingface.co/slperez/LP-Distilled0.1-ONNX
SHA256: 335f98b44e5fab3577b15c3ec53eddd39be228f072d98d7ce11232bcf1c42afd
Original model is split into ordered binary chunks to support static hosts. The browser reassembles and verifies its original SHA256 before execution. Weights are unchanged.
Model card labels the release MIT, experimental, and not for production. It explicitly does not assert compatibility with upstream or dataset licenses. See MODEL_CARD.md. This is a non-commercial feasibility demonstration; no commercial clearance is claimed.

Appearance feature extractor: https://huggingface.co/warmshao/FasterLivePortrait/blob/main/liveportrait_onnx/appearance_feature_extractor.onnx
Upstream code: https://github.com/warmshao/FasterLivePortrait
Expression delta mapping adapted from reference: https://github.com/PowerHouseMan/ComfyUI-AdvancedLivePortrait
Commercial deployment must separately review all model weights, upstream code and training-data terms.

ONNX Runtime Web 1.22.0: https://github.com/microsoft/onnxruntime (MIT; see vendor/LICENSE.txt)
MediaPipe Tasks Vision 0.10.22-rc.20250304: https://github.com/google-ai-edge/mediapipe (Apache-2.0; see mediapipe/LICENSE.txt)
FaceLandmarker model: https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/1/face_landmarker.task

This app sends no photo, image feature, or generated frame to a server. Model files are downloaded from this site's own static assets. No paid API is used.