Wav2Lip gets a GFPGAN facelift inside Automatic1111

What it does This extension drops a full lip-sync pipeline directly into the Automatic1111 interface. It takes a video and an audio track—either uploaded or generated with the built-in Bark text-to-speech—and produces a talking-head clip where the mouth movement matches the speech. The hook is post-processing: instead of serving raw Wav2Lip output, it runs the frames through face-restoration models like CodeFormer and GFPGAN to smooth artifacts around the mouth and skin.

The interesting bit Rather than building a monolithic video model, the project treats A1111 as a deepfake assembly line. It orchestrates Wav2Lip for lip-sync, Bark for voice generation, and FaceFusion for identity swapping, then leans on Stable Diffusion’s existing tooling to patch the visual gaps. It is essentially high-quality glue code for tools you probably already have installed.

Key highlights

• Post-processes Wav2Lip output with CodeFormer and GFPGAN to clean up mouth-region artifacts.
• Bundles experimental Bark TTS with multi-language support, though generated clips are capped at 14 seconds.
• Optional FaceFusion face-swap integration to replace the speaker before syncing lips.
• Supports resuming interrupted generations, and can output a partial video from already-rendered frames.
• Exposes granular mask controls: mouth dilation, face erosion, padding, and blur to fine-tune mouth placement.

Caveats

• Face swap and Bark TTS are explicitly flagged as experimental in the changelog.
• The pipeline hard-crashes if even a single frame in the source video lacks a detectable face.
• Several model weights—Wav2Lip, s3fd, landmark predictors, and the face-swap model—must be downloaded manually and dropped into specific extension subdirectories.

Verdict A convenient addition for Automatic1111 power users who want lip-sync, voice cloning, and face swapping in one tab. If you are not already running A1111, the manual weight setup and WebUI dependency make it a harder sell than standalone alternatives.