Tired of no ComfyUI Lip Sync? None really support Unified mem, AMD etc. Enjoy I made this for AMD but the unified mem community in general

[bkpaine1]

Finally got lip sync working on my Ryzen AI Max+ 395 (Strix Halo) with 64GB unified memory!

Other nodes like DeepFuze fail because they use subprocesses that escape the venv and try to cast bfloat16 tensors directly. Halo-Lipsy fixes this with:

• Native Wav2Lip inference (no subprocess hacks)
• Face detection on CPU (no memory fighting)
• Safe tensor casting for unified memory
• Sync tuning options

Ran it with HunyuanVideo (15GB) still loaded - 4.34 it/s on Wav2Lip!

GitHub: https://github.com/bkpaine1/Halo-Lipsy
ComfyUI Manager PR: Comfy-Org/ComfyUI-Manager#2514

Created with help from Claude Code (Anthropic Opus 4.5). <-- One one day instead of 1 week. NICE Anthropic!