finds.dev← search

// the find

kijai/ComfyUI-LivePortraitKJ

★ 2,189 · Python · MIT · updated Aug 2024

ComfyUI nodes for LivePortrait

A ComfyUI node pack that wraps KwaiVGI's LivePortrait model, letting you animate still images by driving facial motion from a video or webcam feed. Aimed at ComfyUI users who want portrait animation without writing any code.

The MediaPipe path clears the biggest practical hurdle with the original LivePortrait: Insightface's non-commercial license made it a legal liability for anyone building products. Converting pickle weights to safetensors is the right call for both security and compatibility. The ONNX-to-torch landmark conversion that removes the onnxruntime dependency is a nice touch — fewer moving parts in the dependency stack. Near-realtime throughput (~80-100ms) via ComfyUI is genuinely useful for iterative work.

Last push was August 2024 — ten months old at this point, and LivePortrait upstream has kept moving. The README is mostly a changelog with video embeds and no explanation of what the individual nodes actually do or how to connect them; new users are left guessing. The face_alignment subdirectory appears to be a vendored copy of the 1adrianb library rather than a proper dependency, which means security fixes and updates won't flow through automatically. The breaking workflow change from the 'Update' rework with no migration guide is the kind of thing that strands users on an old branch.

View on GitHub →

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →