Builds WAN MultiTalk audio-driven lip-sync talking-avatar video workflows in ComfyUI.
npx skills add https://github.com/artokun/comfyui-mcp --skill wan-multitalk
Please help me install this Agent Skill. Skill: wan-multitalk Source: https://github.com/artokun/comfyui-mcp/tree/main/plugin/skills/wan-multitalk Command: npx skills add https://github.com/artokun/comfyui-mcp --skill wan-multitalk
Creating a talking-avatar or lip-synced video from a still portrait and an audio track requires wiring a complex ComfyUI graph across WAN 2.1 I2V, Wav2Vec speech embeddings, and the MultiTalk head, plus sourcing the correct model files. This Skill provides the complete node graph, model manifest, and parameter guidance to do it correctly.
Use the wan-multitalk skill to build a ComfyUI workflow that animates this portrait image to lip-sync with my audio track.
High-intent search queries and answers about installing and using this skill.
Use the WAN MultiTalk pipeline: load a portrait and audio, isolate the voice with AudioSeparation, generate Wav2Vec embeddings via MultiTalkWav2VecEmbeds, and condition the WAN 2.1 14B I2V sampler through WanVideoImageToVideoMultiTalk. The output is decoded and muxed to mp4 with VHS_VideoCombine.
MultiTalk is audio-driven: it lip-syncs a portrait to a speech track using Wav2Vec embeddings. wan-animate is pose and motion driven, transferring movement from a reference video to a character. Use MultiTalk for talking heads and dubbing, not motion transfer.
The base 480p 10-second configuration needs roughly 8-12 GB VRAM. A low-VRAM variant reaches 6-8.4 GB using more BlockSwap, GGUF quantization, and lower quality settings, while 720p requires about 11-16 GB.
Poor lip-sync usually comes from skipping AudioSeparation on music-heavy audio, so the mouth chases the wrong signal. Also verify the MultiTalk embeds are actually wired into WanVideoImageToVideoMultiTalk and that the audio is not silent after AudioCrop.
No, this graph is single-speaker only. Multi-speaker MultiTalk requires a multi-embed variant that is not included in this pack.
ldzhhxxGenerate images, videos, and audio using ComfyUI workflows.
citedyGenerate and publish branded AI avatar lip-sync video shorts to Instagram Reels and YouTube Shorts.
X-School-AcademyUpscale and enhance video files using the ComfyUI workflow to MP4.
doany-aiRoutes pose-conditioned generation tasks to RunComfy model APIs via CLI inputs and outputs to a user-specified directory.
