comfyui-mcp 0.21.0 → 0.21.1

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "comfyui-mcp",
-  "version": "0.21.0",
+  "version": "0.21.1",
   "mcpName": "io.github.artokun/comfyui-mcp",
   "description": "MCP server + autonomous AI agent for ComfyUI — drive your live graph in natural language from a sidebar agent on Claude OR ChatGPT (your own subscription, no API key). Generate images, video & audio, author and run workflows, manage models and custom nodes. Also a Claude Code plugin with skills, slash commands, and installer packs.",
   "homepage": "https://comfyui-mcp.artokun.io/docs",

package/packs/wan-animate-ofm/install-runpod.sh

@@ -0,0 +1,51 @@
+#!/usr/bin/env bash
+set -euo pipefail
+# ===== Auto-generated by scripts/gen-pack-installers.mjs — do not edit =====
+# Pack: WAN Animate 2.2 — OFM hub (Uni3C + face-swap)  (Linux / RunPod)
+# Run from your ComfyUI root (the folder containing custom_nodes/ and models/).
+[ -d custom_nodes ] || { echo "[ERROR] run from your ComfyUI root (custom_nodes/ not found)"; exit 1; }
+# Resolve the ComfyUI python: its venv first, then portable embed, then \$PYTHON/python3.
+# Installing requirements into the wrong interpreter is why nodes silently fail to load.
+if [ -n "${PYTHON:-}" ]; then PY="$PYTHON";
+elif [ -x ".venv/bin/python" ]; then PY=".venv/bin/python";
+elif [ -x "venv/bin/python" ]; then PY="venv/bin/python";
+elif [ -x "../python_embeded/python" ]; then PY="../python_embeded/python";
+else PY="python3"; fi
+echo "using python: $PY"
+
+clone() { # folder url
+  if [ ! -d "custom_nodes/$1" ]; then
+    echo "  cloning $1"; git clone --depth 1 "$2" "custom_nodes/$1"
+    if [ -f "custom_nodes/$1/requirements.txt" ]; then echo "  installing $1 requirements.txt"; "$PY" -m pip install -r "custom_nodes/$1/requirements.txt"; fi
+  else echo "  $1 present - skip"; fi
+}
+grab() { # relpath url
+  mkdir -p "$(dirname "$1")"
+  if [ ! -f "$1" ]; then echo "  downloading $(basename "$1")"; curl -L -o "$1" "$2"; else echo "  $(basename "$1") present - skip"; fi
+}
+
+echo "-------- custom nodes --------"
+clone "ComfyUI-Manager" "https://github.com/ltdrdata/ComfyUI-Manager.git"
+clone "ComfyUI-WanVideoWrapper" "https://github.com/kijai/ComfyUI-WanVideoWrapper"
+clone "ComfyUI-KJNodes" "https://github.com/kijai/ComfyUI-KJNodes"
+clone "ComfyUI-WanAnimatePreprocess" "https://github.com/kijai/ComfyUI-WanAnimatePreprocess"
+clone "ComfyUI-segment-anything-2" "https://github.com/kijai/ComfyUI-segment-anything-2"
+clone "ComfyUI-VideoHelperSuite" "https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite"
+clone "ComfyMath" "https://github.com/evanspearman/ComfyMath"
+clone "CRT-Nodes" "https://github.com/plugcrypt/CRT-Nodes"
+
+echo "-------- models --------"
+grab "models/diffusion_models/WanModel.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/resolve/main/Wan22Animate/Wan2_2-Animate-14B_fp8_scaled_e4m3fn_KJ_v2.safetensors"
+grab "models/vae/vae.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Wan2_1_VAE_bf16.safetensors"
+grab "models/clip_vision/klip_vision.safetensors" "https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/clip_vision/clip_vision_h.safetensors"
+grab "models/text_encoders/text_enc.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/umt5-xxl-enc-fp8_e4m3fn.safetensors"
+grab "models/controlnet/Wan21_Uni3C_controlnet_fp16.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Wan21_Uni3C_controlnet_fp16.safetensors"
+grab "models/loras/light.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors"
+grab "models/loras/wan.reworked.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/LoRAs/Wan22_relight/WanAnimate_relight_lora_fp16.safetensors"
+grab "models/loras/WanPusa.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Pusa/Wan21_PusaV1_LoRA_14B_rank512_bf16.safetensors"
+grab "models/loras/WanFun.reworked.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/LoRAs/Wan22_FunReward/Wan2.2-Fun-A14B-InP-LOW-MPS_resized_dynamic_avg_rank_22_bf16.safetensors"
+grab "models/detection/yolov10m.onnx" "https://huggingface.co/Wan-AI/Wan2.2-Animate-14B/resolve/main/process_checkpoint/det/yolov10m.onnx"
+grab "models/detection/vitpose_h_wholebody_model.onnx" "https://huggingface.co/Kijai/vitpose_comfy/resolve/main/onnx/vitpose_h_wholebody_model.onnx"
+grab "models/detection/vitpose_h_wholebody_data.bin" "https://huggingface.co/Kijai/vitpose_comfy/resolve/main/onnx/vitpose_h_wholebody_data.bin"
+
+echo "DONE. Restart ComfyUI, then load workflow.json."

package/packs/wan-animate-ofm/install-windows.bat

@@ -0,0 +1,56 @@
+@echo off
+setlocal EnableExtensions EnableDelayedExpansion
+chcp 65001 >nul
+rem ===== Auto-generated by scripts/gen-pack-installers.mjs — do not edit =====
+rem Pack: WAN Animate 2.2 — OFM hub (Uni3C + face-swap)
+rem Run from your ComfyUI root (the folder containing custom_nodes\ and models\).
+if not exist "custom_nodes" ( echo [ERROR] Run from your ComfyUI root ^(custom_nodes\ not found^). & pause & exit /b 1 )
+where git >nul 2>&1 || ( echo [ERROR] git not found in PATH. & pause & exit /b 1 )
+where curl >nul 2>&1 || ( echo [ERROR] curl not found in PATH. & pause & exit /b 1 )
+rem Resolve the ComfyUI python: venv first, then portable embed, then PATH python.
+set "PY=%CD%\.venv\Scripts\python.exe"
+if not exist "%PY%" set "PY=%CD%\venv\Scripts\python.exe"
+if not exist "%PY%" set "PY=%CD%\..\python_embeded\python.exe"
+if not exist "%PY%" set "PY=python"
+echo using python: %PY%
+
+echo -------- custom nodes --------
+call :clone "ComfyUI-Manager" "https://github.com/ltdrdata/ComfyUI-Manager.git"
+call :clone "ComfyUI-WanVideoWrapper" "https://github.com/kijai/ComfyUI-WanVideoWrapper"
+call :clone "ComfyUI-KJNodes" "https://github.com/kijai/ComfyUI-KJNodes"
+call :clone "ComfyUI-WanAnimatePreprocess" "https://github.com/kijai/ComfyUI-WanAnimatePreprocess"
+call :clone "ComfyUI-segment-anything-2" "https://github.com/kijai/ComfyUI-segment-anything-2"
+call :clone "ComfyUI-VideoHelperSuite" "https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite"
+call :clone "ComfyMath" "https://github.com/evanspearman/ComfyMath"
+call :clone "CRT-Nodes" "https://github.com/plugcrypt/CRT-Nodes"
+
+echo -------- models --------
+call :grab "models\diffusion_models\WanModel.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/resolve/main/Wan22Animate/Wan2_2-Animate-14B_fp8_scaled_e4m3fn_KJ_v2.safetensors"
+call :grab "models\vae\vae.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Wan2_1_VAE_bf16.safetensors"
+call :grab "models\clip_vision\klip_vision.safetensors" "https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/clip_vision/clip_vision_h.safetensors"
+call :grab "models\text_encoders\text_enc.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/umt5-xxl-enc-fp8_e4m3fn.safetensors"
+call :grab "models\controlnet\Wan21_Uni3C_controlnet_fp16.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Wan21_Uni3C_controlnet_fp16.safetensors"
+call :grab "models\loras\light.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors"
+call :grab "models\loras\wan.reworked.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/LoRAs/Wan22_relight/WanAnimate_relight_lora_fp16.safetensors"
+call :grab "models\loras\WanPusa.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Pusa/Wan21_PusaV1_LoRA_14B_rank512_bf16.safetensors"
+call :grab "models\loras\WanFun.reworked.safetensors" "https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/LoRAs/Wan22_FunReward/Wan2.2-Fun-A14B-InP-LOW-MPS_resized_dynamic_avg_rank_22_bf16.safetensors"
+call :grab "models\detection\yolov10m.onnx" "https://huggingface.co/Wan-AI/Wan2.2-Animate-14B/resolve/main/process_checkpoint/det/yolov10m.onnx"
+call :grab "models\detection\vitpose_h_wholebody_model.onnx" "https://huggingface.co/Kijai/vitpose_comfy/resolve/main/onnx/vitpose_h_wholebody_model.onnx"
+call :grab "models\detection\vitpose_h_wholebody_data.bin" "https://huggingface.co/Kijai/vitpose_comfy/resolve/main/onnx/vitpose_h_wholebody_data.bin"
+
+echo DONE. Restart ComfyUI, then load workflow.json.
+pause
+exit /b
+
+:clone
+if not exist "custom_nodes\%~1" (
+  echo   cloning %~1
+  git clone --depth 1 "%~2" "custom_nodes\%~1"
+  if exist "custom_nodes\%~1\requirements.txt" ( echo   installing %~1 requirements.txt & "%PY%" -m pip install -r "custom_nodes\%~1\requirements.txt" )
+) else ( echo   %~1 present - skip )
+goto :eof
+
+:grab
+if not exist "%~dp1" mkdir "%~dp1"
+if not exist "%~1" ( echo   downloading %~nx1 & curl -L --ssl-no-revoke -o "%~1" "%~2" ) else ( echo   %~nx1 present - skip )
+goto :eof

package/packs/wan-animate-ofm/manifest.yaml

@@ -0,0 +1,78 @@
+# WAN Animate 2.2 — "OFM hub" variant. PERSONAL pack (see pack.yaml notes).
+# Decoded + flattened from a teskor-hub WAN Animate hub workflow: drive a
+# character image with a video (ViTPose+YOLO pose/face), Uni3C controlnet for
+# temporal stability, optional bypassed SAM2 face-swap masking, color-match,
+# save video.
+#
+# MODEL NAMING: the source workflow uses short PLACEHOLDER filenames
+# (WanModel/vae/klip_vision/text_enc + light/wan.reworked/WanPusa/WanFun.reworked).
+# Each entry below downloads the REAL Kijai file but saves it UNDER THE PLACEHOLDER
+# NAME the graph expects, so the workflow loads without edits.
+#
+# PRIVATE NODES (NOT auto-installable — install manually): this graph also uses
+# four teskor-hub nodes whose repos are private:
+#   TSColorMatch, TSVideoCombineNoMetadata, TSPreviewImageNoMetadata, TSPoseDataSmoother
+#   (github.com/teskor-hub/NEW-UTILS, github.com/teskor-hub/comfyui-teskors-utils)
+# Without them the graph won't fully load. Swap to VHS_VideoCombine / KJNodes
+# ColorMatch / core PreviewImage if you don't have the teskor packs.
+
+pip: []
+
+custom_nodes:
+  - https://github.com/ltdrdata/ComfyUI-Manager.git
+  - https://github.com/kijai/ComfyUI-WanVideoWrapper        # WanVideo* (model/vae/lora/sampler/decode/animate/uni3c)
+  - https://github.com/kijai/ComfyUI-KJNodes               # DrawMaskOnImage, BlockifyMask, GrowMaskWithBlur, ImageResizeKJv2, GetImageSizeAndCount
+  - https://github.com/kijai/ComfyUI-WanAnimatePreprocess  # PoseAndFaceDetection, OnnxDetectionModelLoader (vitpose+yolov10m auto-download)
+  - https://github.com/kijai/ComfyUI-segment-anything-2    # DownloadAndLoadSAM2Model, Sam2Segmentation (bypassed face-swap branch)
+  - https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite # VHS_LoadVideo
+  - https://github.com/evanspearman/ComfyMath              # CM_FloatToInt (bogus aux_id in the file said Workflow-Encrypt — it's ComfyMath)
+  - https://github.com/plugcrypt/CRT-Nodes                 # SimpleKnobNode (bypassed branch)
+  # --- PRIVATE / user-supplied (NOT cloned) — see header note ---
+  # - https://github.com/teskor-hub/NEW-UTILS               # TSColorMatch
+  # - https://github.com/teskor-hub/comfyui-teskors-utils   # TSPoseDataSmoother, TSVideoCombineNoMetadata, TSPreviewImageNoMetadata
+
+models:
+  # --- WAN 2.2 Animate 14B (fp8 scaled, Kijai v2) -> placeholder WanModel ---
+  - url: https://huggingface.co/Kijai/WanVideo_comfy_fp8_scaled/resolve/main/Wan22Animate/Wan2_2-Animate-14B_fp8_scaled_e4m3fn_KJ_v2.safetensors
+    local_path: diffusion_models/WanModel.safetensors
+
+  # --- WAN 2.1 VAE -> placeholder vae ---
+  - url: https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Wan2_1_VAE_bf16.safetensors
+    local_path: vae/vae.safetensors
+
+  # --- CLIP vision H -> placeholder klip_vision (Comfy-Org repackaged; standard) ---
+  - url: https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/clip_vision/clip_vision_h.safetensors
+    local_path: clip_vision/klip_vision.safetensors
+
+  # --- umt5-xxl text encoder (fp8) -> placeholder text_enc ---
+  - url: https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/umt5-xxl-enc-fp8_e4m3fn.safetensors
+    local_path: text_encoders/text_enc.safetensors
+
+  # --- Uni3C controlnet (temporal stability) — real name kept ---
+  - url: https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Wan21_Uni3C_controlnet_fp16.safetensors
+    local_path: controlnet/Wan21_Uni3C_controlnet_fp16.safetensors
+
+  # --- LoRA stack (WanVideoLoraSelectMulti) -> placeholders ---
+  #     light@1.0 = lightx2v 4-step distill
+  - url: https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Lightx2v/lightx2v_I2V_14B_480p_cfg_step_distill_rank64_bf16.safetensors
+    local_path: loras/light.safetensors
+  #     wan.reworked@0.3 = WanAnimate relight LoRA
+  - url: https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/LoRAs/Wan22_relight/WanAnimate_relight_lora_fp16.safetensors
+    local_path: loras/wan.reworked.safetensors
+  #     WanPusa@0.9 = Pusa V1 temporal LoRA (single-file Wan2.1)
+  - url: https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/Pusa/Wan21_PusaV1_LoRA_14B_rank512_bf16.safetensors
+    local_path: loras/WanPusa.safetensors
+  #     WanFun.reworked@0.5 = Wan2.2-Fun-A14B-InP LoRA (resized rank-22)
+  - url: https://huggingface.co/Kijai/WanVideo_comfy/resolve/main/LoRAs/Wan22_FunReward/Wan2.2-Fun-A14B-InP-LOW-MPS_resized_dynamic_avg_rank_22_bf16.safetensors
+    local_path: loras/WanFun.reworked.safetensors
+
+  # --- Pose detection ONNX (OnnxDetectionModelLoader loads from models/detection/) ---
+  #     YOLOv10m (from the official Wan2.2-Animate checkpoint)
+  - url: https://huggingface.co/Wan-AI/Wan2.2-Animate-14B/resolve/main/process_checkpoint/det/yolov10m.onnx
+    local_path: detection/yolov10m.onnx
+  #     ViTPose-Huge wholebody — split into model + data.bin (ONNX 2GB limit); both
+  #     must sit in the same folder, the .onnx is the one selected in the loader.
+  - url: https://huggingface.co/Kijai/vitpose_comfy/resolve/main/onnx/vitpose_h_wholebody_model.onnx
+    local_path: detection/vitpose_h_wholebody_model.onnx
+  - url: https://huggingface.co/Kijai/vitpose_comfy/resolve/main/onnx/vitpose_h_wholebody_data.bin
+    local_path: detection/vitpose_h_wholebody_data.bin

package/packs/wan-animate-ofm/pack.yaml

@@ -0,0 +1,43 @@
+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.
+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.