comfyui-mcp/packs/wan-multitalk/pack.yaml

Local-first, agent-native control plane for ComfyUI — MCP server + autonomous sidebar agent that drives your live graph in natural language on ANY LLM: Claude/ChatGPT/Gemini on your subscription (no API key), free local models via Ollama (fully offline),

name: wan-multitalk
display_name: WAN MultiTalk — audio-driven talking avatar
family: wan
kind: talking-avatar
description: >-
  Audio-driven talking-avatar / lip-sync video. MeiGen-AI MultiTalk on WAN 2.1 14B
  I2V (kijai WanVideoWrapper): drive a still portrait's lip-sync + head motion from
  an audio track. 480p, ~10s. Wav2Vec speech embeddings condition the WAN sampler;
  the lightx2v step-distill LoRA keeps it to a few steps.
vram: "12GB+"
workflow: workflow.json
skill: wan-multitalk
launch_args: []
sources:
  model: https://huggingface.co/Kijai/WanVideo_comfy
  multitalk: https://huggingface.co/MeiGen-AI/MeiGen-MultiTalk
  gguf: https://huggingface.co/city96/Wan2.1-I2V-14B-480P-gguf
notes:
  - Run the generated installer from your ComfyUI root.
  - Needs kijai's ComfyUI-WanVideoWrapper (all WanVideo*/MultiTalk* nodes), KJNodes
    (ImageResizeKJv2), VideoHelperSuite (VHS_VideoCombine), and
    christian-byrne's audio-separation-nodes-comfyui (AudioSeparation + AudioCrop).
  - Uses umt5_xxl_fp16.safetensors because the bundled WanVideoWrapper loader
    rejects the scaled_fp8 UMT5 checkpoint; do not substitute generic t5xxl weights.
  - Wav2Vec is auto-downloaded on first run by DownloadAndLoadWav2VecModel — no
    manual model needed for it.
  - VERIFY the model URLs against your setup before a clean install — the GGUF and
    a couple of kijai filenames vary by mirror.
post_install:
  - Restart ComfyUI, load workflow.json.
  - LoadImage = a front-facing portrait; LoadAudio = the speech track. Queue.