Lip Forcing 🗣️

Few-Step Autoregressive Diffusion for Real-time Lip Synchronization  ·  14B student  ·  Paper  ·  Project  ·  Code  ·  Weights

Give it a talking-head video and a driving audio clip — it detects and aligns the face, then regenerates the mouth to match the audio with a 2-step causal diffusion student. Audio is capped to the first few seconds per run.