comfyui-mcp/packs/wan-animate-ofm/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-animate-ofm
display_name: WAN Animate 2.2 — OFM hub (Uni3C + face-swap)
family: wan
kind: video
description: >-
  WAN Animate 2.2 video-to-video character animation — the "OFM hub" variant,
  decoded and flattened from a teskor-hub workflow. Drives a character image with
  a driving video (ViTPose + YOLOv10 pose/face detection), adds Uni3C controlnet
  for temporal stability, color-matches, and saves video. A SAM2 face-swap /
  deep-fake masking branch ships BYPASSED (opt-in). Runs on the Kijai WanVideo
  stack (Wan2.2-Animate-14B fp8 + Wan2.1 VAE + clip_vision_h + umt5) with a LoRA
  stack of lightx2v (4-step distill), WanAnimate relight, Pusa V1 temporal, and
  Wan2.2-Fun. Distinct from the SeC-based `wan-animate` pack. For the
  maintained/recommended character-animation pipeline, prefer the `artokun-flow` pack.
vram: 24GB+
workflow: workflow.json
skill: video-extend
launch_args: []
sources:
  kijai: https://huggingface.co/Kijai/WanVideo_comfy
  kijai_fp8: https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled
notes:
  - PERSONAL PACK — requires FOUR private teskor-hub nodes that are NOT
    auto-installable (TSColorMatch, TSVideoCombineNoMetadata,
    TSPreviewImageNoMetadata, TSPoseDataSmoother). Install them manually, or swap
    to standard equivalents (VHS_VideoCombine / KJNodes ColorMatch / core
    PreviewImage) — without them the graph won't fully load.
  - Models download under the workflow's PLACEHOLDER names (WanModel / vae /
    klip_vision / text_enc + light / wan.reworked / WanPusa / WanFun.reworked) so
    the graph loads without edits — see manifest.yaml for the real Kijai files.
  - It IS a Kijai stack end-to-end (model in WanVideo_comfy_fp8_scaled, the rest +
    LoRAs in WanVideo_comfy). Exact LoRA variant picks (rank/precision) were
    inferred from the canonical WAN_ANIMATE_ULTRA workflow + the video-extend
    skill (Pusa); confirm against your own files if a rank differs.
  - The SAM2 "Face swap" branch (DownloadAndLoadSAM2Model -> Sam2Segmentation ->
    mask nodes) ships BYPASSED — un-bypass those 7 nodes to enable masked
    face-swap. SAM2 + the pose ONNX models (vitpose/yolov10m) auto-download via
    their nodes.
  - NOT render-verified here (WAN Animate models are ~tens of GB and weren't
    installed). Static-validated only (clean flatten + structural lint).
post_install:
  - Install the private teskor-hub nodes (or swap to standard equivalents).
  - Restart ComfyUI (or ComfyUI-Manager -> Install missing custom nodes).
  - Load workflow.json.