MuseTalk (GPU lip-sync)

High-quality generative lip-sync (visual dubbing) via MuseTalk. Requires an NVIDIA GPU (CUDA) and Python 3.10+. Small download: the PyTorch runtime and multi-GB models install on first use. MIT-licensed and commercial-OK.

Options

In Settings › Add-ons › MuseTalk (GPU lip-sync) › Configure.

MuseTalk version

Default: v15

MuseTalk model version. 1.5 gives sharper mouths; 1.0 is the original.

Bbox shift

Default: 0

Shifts the face crop vertically to tune mouth openness. Positive opens the mouth more, negative closes it. MuseTalk 1.0 only.

Extra margin

Default: 10

Extra pixels added below the face crop (jaw region). MuseTalk 1.5 only.

Batch size

Default: 8

Frames generated per UNet forward pass. Higher is faster but needs more VRAM; lower it on 'oom' errors.

Output FPS

Default: 25

Frame rate of the generated video. MuseTalk is trained at 25 fps; leave at 25 for best results.

Use float16

Default: on

Run the UNet/VAE in fp16 to cut VRAM use. Disable if you see numerical artifacts.

Platforms

System	Download
Linux x86-64	17.7 MB
Windows x86-64	8.6 MB

License

Add-on

MIT

MuseTalk (GPU lip-sync)

Options

Platforms

License

Versions

Other Lip-sync add-ons