ComfyUI Node
H3 Lip Sync Wav2Lip (phoneme-level mouth)
A ComfyUI node in minimax/h3/audio with 11 inputs and 3 outputs.
H3 Lip Sync Wav2Lip (phoneme-level mouth)
- images
- audio
- frames
- audio
- report
◄fps24.000►
◄checkpointwav2lip_gan.pth►
◄pads0 10 0 0►
◄resize_factor1►
◄nosmoothfalse►
◄audio_offset_frames0►
◄python_exeG:\lipsync-tools\venv-latentsync\Scripts\python.exe►
◄wav2lip_dirG:\lipsync-tools\Wav2Lip►
◄ffmpeg_dirF:\ffmpeg\ffmpeg-8.0.1-essentials_build\bin►
Categoryminimax/h3/audio
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| images | IMAGE | Frames to re-synthesise. Only the mouth is replaced; face, hair and body pixels are kept. | |
| audio | AUDIO | The soundtrack to lip-sync to. Returned (optionally shifted) for the mux. | |
| fps | FLOAT | 24.0001–120 | — |
| checkpoint | COMBO | wav2lip_gan.pth | wav2lip_gan = sharper picture; wav2lip = slightly tighter lipsync and a softer mouth (both measured on this machine). |
| pads | STRING | 0 10 0 0 | Extra crop around the detected face: top bottom left right, in pixels. |
| resize_factor | INT | 11–4 | Downscale before processing (1 = full resolution). 2 is faster on big frames. |
| nosmooth | BOOLEAN | false | Disable the temporal smoothing of the detected face box. |
| audio_offset_framesopt | INT | 0-48–48 | Shifts the returned audio against the frames (same meaning as the driving nodes). Leave 0 to lip-sync in place. |
| python_exeopt | STRING | G:\lipsync-tools\venv-latentsync\Scripts\python.exe | Interpreter that can import Wav2Lip's deps (librosa, torch, face_detection). |
| wav2lip_diropt | STRING | G:\lipsync-tools\Wav2Lip | Wav2Lip checkout containing inference.py and checkpoints/. |
| ffmpeg_diropt | STRING | F:\ffmpeg\ffmpeg-8.0.1-essentials_build\bin | Directory prepended to PATH; Wav2Lip shells out to ffmpeg for audio. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| frames | IMAGE | — |
| audio | AUDIO | — |
| report | STRING | — |