feat: add virtual environment support for ComfyUI

Changes:
- Create venv for ComfyUI in setup/comfyui-base script
- Install extension dependencies: GitPython, opencv-python-headless,
  diffusers, insightface, onnxruntime
- Update start.sh to activate venv before running
- Add musicgen model directory

This fixes import errors for custom nodes:
- ComfyUI-Manager (needs GitPython)
- ComfyUI-Impact-Pack (needs opencv)
- ComfyUI-VideoHelperSuite (needs opencv)
- ComfyUI-CogVideoXWrapper (needs diffusers)
- ComfyUI-Inspire-Pack (needs insightface, onnxruntime)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

arty.yml

@@ -229,19 +229,34 @@ scripts:
       cd $COMFYUI_ROOT && git pull
     fi
 
+    # Create virtual environment for ComfyUI
+    echo "Creating virtual environment for ComfyUI..."
+    cd $COMFYUI_ROOT
+    if [ ! -d "venv" ]; then
+      python3 -m venv venv
+    fi
+    source venv/bin/activate
+
     # Install ComfyUI dependencies
     echo "Installing ComfyUI dependencies..."
-    sudo pip3 install -r $COMFYUI_ROOT/requirements.txt
+    pip install -r requirements.txt
 
     # Install additional ComfyUI dependencies
-    if [ -f "comfyui/requirements.txt" ]; then
+    if [ -f "$AI_ROOT/comfyui/requirements.txt" ]; then
       echo "Installing additional ComfyUI dependencies..."
-      sudo pip3 install -r comfyui/requirements.txt
+      pip install -r $AI_ROOT/comfyui/requirements.txt
     fi
 
+    # Install common extension dependencies
+    echo "Installing common extension dependencies..."
+    pip install GitPython opencv-python-headless diffusers insightface onnxruntime
+
+    deactivate
+    cd $AI_ROOT
+
     # Create model directory structure
     echo "Creating ComfyUI model directories..."
-    mkdir -p $COMFYUI_ROOT/models/{checkpoints,unet,vae,loras,clip,clip_vision,controlnet,ipadapter,embeddings,upscale_models,video_models,animatediff_models,animatediff_motion_lora,audio_models,configs,diffusers,diffusion_models}
+    mkdir -p $COMFYUI_ROOT/models/{checkpoints,unet,vae,loras,clip,clip_vision,controlnet,ipadapter,embeddings,upscale_models,video_models,animatediff_models,animatediff_motion_lora,audio_models,configs,diffusers,diffusion_models,musicgen}
 
     # Create symlink to huggingface cache
     echo "Creating symlink to HuggingFace cache..."

comfyui/start.sh

@@ -16,12 +16,20 @@ export TQDM_DISABLE=1
 # Navigate to ComfyUI directory
 cd "${COMFYUI_DIR}" || exit 1
 
+# Activate virtual environment
+if [ -d "venv" ]; then
+  echo "Activating ComfyUI virtual environment..."
+  source venv/bin/activate
+else
+  echo "WARNING: venv not found, using system Python"
+fi
+
 echo "Starting ComfyUI on port 8188..."
 echo "Access at: http://localhost:8188"
 echo "Using HuggingFace cache: ${HF_CACHE}"
 
 # Start ComfyUI with GPU support
-python3 main.py \
+python main.py \
   --listen 0.0.0.0 \
   --port 8188 \
   --enable-cors-header \