ComfyUI_Sonic

Why we included this project

Animators and video editors working with talking-head portraits can run Sonic as a ComfyUI custom node rather than a standalone script, so it slots into an existing node graph next to other image and video nodes. The method takes a voice track and drives the portrait's lip motion, head pose, and expressions from global audio cues. Setup follows the usual custom-node pattern: clone into ComfyUI/custom_node, install requirements, then drop the Sonic weights, a Whisper-tiny model, and an SVD checkpoint into the expected folders. The maintainer has also fixed a batch-mismatch bug at non-25 fps, added a duration control in seconds, and lowered first-run memory pressure on 12 GB GPUs with Mac MPS support, which makes it friendlier for mid-range hardware than the upstream research code.