Neural Video Compression

150-250x smaller video with GFVC (Generative Face Video Coding)

Uses MediaPipe Face Landmarker (~5MB)

GFVC compresses video by extracting a single keyframe and per-frame facial motion data.

Instead of encoding every pixel, only face landmarks and expressions are stored.

The receiver uses a face animation model to reconstruct the video from this data.

Audio Compression | Messaging Demo | MediaPipe Docs