ComfyUI-LatentSyncWrapper

Why we included this project

Lip-sync usually means stepping out of ComfyUI to run a separate pipeline, so getting it as a plain custom node inside the graph is the main draw here. The wrapper sits on ByteDance's LatentSync: drop in a talking-head clip, give it an audio track, and the node redraws the mouth to match the speech, with knobs for lip-movement strength and denoising steps. The 1.6 model was trained at 512x512 resolution, which clears up the mushy teeth and lip detail that hurt earlier releases, and the implementation keeps GPU memory low enough to work on a 20GB card. One thing to plan around: the LatentSync 1.6 weights live in a gated HuggingFace repository, so checkpoints have to be fetched by hand, and the plugin demands a clean reinstall rather than an in-place update.