This repository provides fully automated batch scripts to install and launch ComfyUI on Windows, optimized for both Intel Arc GPUs and Intel Ultra Core iGPUs (Meteor Lake/Core Ultra series) using the XPU backend.
You get:
- The latest ComfyUI and official frontend
- An isolated Python virtual environment (venv)
- Intel Arc/XPU-optimized PyTorch
- No need to manually patch device code
- Optional batch to install ComfyUI Manager
- No dependency conflicts: ComfyUI runs in its own venv, separate from AI Playground, Pinokio, or other tools.
- Always up-to-date: Scripts fetch the latest ComfyUI and frontend versions.
- Automatic XPU support: PyTorch XPU is installed and used by default.
- No manual patching: No need to edit
model_management.pyfor device selection. - ComfyUI Manager install script included.
... This setup is tested and confirmed to work with:
| GPU Type | Supported | Notes |
|---|---|---|
| Intel Arc (A-Series) | âś… Yes | Full support with PyTorch XPU. (e.g. Arc A770, A750, A580, A380, A310) |
| Intel Arc Pro (Workstation) | âś… Yes | Same as above. |
| Intel Ultra Core iGPU | âś… Yes | Supported via PyTorch XPU (e.g. Intel Core Ultra 5/7/9 Meteor Lake NPU/iGPU). |
| Intel Iris Xe (integrated) | Experimental, limited or no support in current PyTorch XPU builds. May fallback to CPU. | |
| NVIDIA (GTX/RTX) | âś… Yes | Use the official CUDA/Windows portable or conda install. |
| AMD Radeon (RDNA/ROCm) | ROCm support is limited and not recommended for most users. | |
| CPU only | âś… Yes | Works, but extremely slow for image/video generation. |
Intel Ultra Core iGPU Support:
- Intel Ultra Core (Meteor Lake, "Core Ultra 5/7/9" etc.) iGPUs are supported with this guide, as they use the same PyTorch XPU backend as Intel Arc discrete GPUs.
- You do not need a discrete Arc GPU; the integrated GPU in Intel Ultra Core CPUs will be used if present and drivers are up to date.
- Performance will be lower than Arc A-series, but you get full node-based ComfyUI functionality. ... Intel Iris Xe and UHD Graphics:
- Intel Iris Xe iGPU support is experimental. Some features may not work or may fall back to CPU.
- Intel UHD Graphics (older iGPUs) are not supported for AI acceleration, and ComfyUI will use CPU only.
- Windows 10/11
- Intel Arc GPU (A-series, Arc Pro) or Intel Ultra Core iGPU (Meteor Lake/Core Ultra series)
- Python 3.10 or 3.11 (download here)
- Git for Windows (download here)
- At least 50GB free disk space (for models, nodes, outputs, etc.)
- Latest Intel Graphics drivers
- Backup your
models,custom_nodes, andworkflowsfolders if you have a previous ComfyUI installation. - Delete or rename any old
C:\ComfyUIfolder if you want a completely clean install.
... Save the following as install_comfyui_venv.bat and run it (double-click or right-click > Run as administrator):
@echo off
REM === [OPTIONAL] Delete old ComfyUI folder for a clean install ===
REM rmdir /s /q C:\ComfyUI
REM 1. Clone the latest ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git C:\ComfyUI
REM 2. Go to the ComfyUI directory
cd /d C:\ComfyUI
REM 3. Create an isolated Python venv
python -m venv comfyui_venv
REM 4. Activate the venv
call comfyui_venv\Scripts\activate.bat
REM 5. Upgrade pip
python -m pip install --upgrade pip
REM 6. Install base dependencies
pip install -r requirements.txt
REM 7. Uninstall any previous torch
pip uninstall -y torch torchvision torchaudio
REM 8. Install Intel Arc (XPU) optimized torch
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/xpu
REM 9. Upgrade to the latest official ComfyUI frontend
pip install --upgrade comfyui-frontend-package
... REM 10. Show torch version for confirmation
python -c "import torch; print('Torch version:', torch.__version__)"
echo.
echo ==========================
echo Installation complete!
echo If you see Torch version ...+xpu, you are ready.
echo You can now launch ComfyUI with the start batch file.
echo ==========================
pauseSave the following as start_comfyui_venv.bat and use it to start ComfyUI every time:
@echo off
set SYCL_CACHE_PERSISTENT=1
cd /d C:\ComfyUI
call comfyui_venv\Scripts\activate.bat
REM Launch with always up-to-date official frontend
python main.py --preview-method auto --output-directory C:\Users\%USERNAME%\Documents\AI-Playground\media --lowvram --disable-ipex-optimize --bf16-unet --reserve-vram 6.0 --front-end-version Comfy-Org/ComfyUI_frontend@latest
pauseSave this as install_comfyui_manager_venv.bat and run it after installing ComfyUI:
@echo off... REM Activate the venv
cd /d C:\ComfyUI
call comfyui_venv\Scripts\activate.bat
REM Go to custom_nodes directory
cd custom_nodes
REM Clone ComfyUI-Manager (if not already present)
IF NOT EXIST comfyui-manager (
git clone https://github.com/ltdrdata/ComfyUI-Manager comfyui-manager
) ELSE (
echo comfyui-manager already exists, skipping clone.
)
REM Install ComfyUI-Manager dependencies
pip install -r comfyui-manager\requirements.txt
echo.
echo ==========================
echo ComfyUI Manager installed!
echo Restart ComfyUI to enable the Manager.
echo ==========================
pauseSave this as install_sageattention_compatibility.bat and run it if you need SageAttention compatibility:
@echo off
REM Activate the venv
cd /d C:\ComfyUI
call comfyui_venv\Scripts\activate.bat
echo ====================================
echo Installing SageAttention compatibility layer for Intel XPU
echo ====================================
REM Create the proper directory structure
cd custom_nodes
IF EXIST sageattention (
rd /s /q sageattention
)
mkdir sageattention
REM Create a single __init__.py file with all the necessary code
echo import torch > sageattention\__init__.py
echo from torch import nn >> sageattention\__init__.py
echo. >> sageattention\__init__.py
echo class SageAttention(nn.Module): >> sageattention\__init__.py
echo def __init__(self, *args, **kwargs): >> sageattention\__init__.py
echo super().__init__() >> sageattention\__init__.py
echo def forward(self, x): >> sageattention\__init__.py
echo return x >> sageattention\__init__.py
echo. >> sageattention\__init__.py
echo def sageattn(*args, **kwargs): >> sageattention\__init__.py
echo return None >> sageattention\__init__.py
echo. >> sageattention\__init__.py
echo class SageAttentionNode: >> sageattention\__init__.py
echo @classmethod >> sageattention\__init__.py
echo def INPUT_TYPES(s): >> sageattention\__init__.py
echo return {"required": {"tensor": ("TENSOR",)}} >> sageattention\__init__.py
echo RETURN_TYPES = ("TENSOR",) >> sageattention\__init__.py
echo FUNCTION = "forward" >> sageattention\__init__.py
echo CATEGORY = "advanced" >> sageattention\__init__.py
echo. >> sageattention\__init__.py
echo def forward(self, tensor): >> sageattention\__init__.py
echo return (tensor,) >> sageattention\__init__.py
echo. >> sageattention\__init__.py
echo NODE_CLASS_MAPPINGS = { >> sageattention\__init__.py
echo "SageAttention": SageAttentionNode >> sageattention\__init__.py
echo } >> sageattention\__init__.py
echo. >> sageattention\__init__.py
echo print("SageAttention XPU compatibility layer loaded") >> sageattention\__init__.py
echo.
echo ====================================
echo SageAttention compatibility layer installed!
echo This allows ComfyUI to run without errors when SageAttention is requested.
echo Note: This does not provide actual SageAttention functionality.
echo ====================================
pauseSave this as update_comfyUI_xpu_intel.bat and run it to Update:
@echo off
echo ===== ComfyUI Installer/Updater =====
REM Check if ComfyUI is already installed
IF NOT EXIST "C:\ComfyUI" (
REM === Original installation script ===
REM === [OPTIONAL] Remove old ComfyUI folder for a clean install ===
REM rmdir /s /q C:\ComfyUI
REM 1. Clone the latest version of ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git C:\ComfyUI
REM 2. Go to ComfyUI folder
cd /d C:\ComfyUI
REM 3. Create an isolated Python venv
python -m venv comfyui_venv
REM 4. Activate the venv
call comfyui_venv\Scripts\activate.bat
REM 5. Update pip
python -m pip install --upgrade pip
REM 6. Install base dependencies
pip install -r requirements.txt
REM 7. Uninstall any previous torch version
pip uninstall -y torch torchvision torchaudio
REM 8. Install Intel Arc optimized torch (XPU)
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/xpu
REM 9. Update ComfyUI frontend to the latest official version (optional)
pip install --upgrade comfyui-frontend-package
REM 10. Check installed torch version (should display +xpu)
python -c "import torch; print('Torch version:', torch.__version__)"
echo.
echo ==========================
echo Installation complete!
echo If torch version shows +xpu, you're ready to go.
echo You can now launch ComfyUI with the startup batch.
echo ==========================
) ELSE (
REM === Update script ===
echo ComfyUI is already installed. Updating...
REM 1. Go to ComfyUI folder
cd /d C:\ComfyUI
REM 2. Activate the venv
call comfyui_venv\Scripts\activate.bat
REM 3. Pull the latest changes from GitHub
echo Updating ComfyUI core...
git pull
REM 4. Update pip
echo Updating pip...
python -m pip install --upgrade pip
REM 5. Update dependencies
echo Updating dependencies...
pip install -r requirements.txt
REM 6. Update torch (keeping Intel XPU version)
echo Updating PyTorch...
pip install --upgrade --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/xpu
REM 7. Update frontend to latest version
echo Updating ComfyUI frontend...
pip install --upgrade comfyui-frontend-package
REM 8. Check torch version
echo Checking torch version...
python -c "import torch; print('Torch version:', torch.__version__)"
echo.
echo ==========================
echo Update complete!
echo If torch version shows +xpu, you're ready to go.
echo ==========================
)
pauseSome workflows and custom nodes require SageAttention, which is primarily designed for NVIDIA GPUs with CUDA support. For Intel XPU users, this repository includes a compatibility layer:
- Run
install_sageattention_compatibility.batto install a lightweight compatibility layer - This allows workflows that expect SageAttention to run without errors
- Note that this is a pass-through implementation that doesn't provide the actual attention mechanism
- The compatibility layer prevents errors while still leveraging Intel XPU for the rest of the pipeline
This approach is particularly useful when working with workflows designed for NVIDIA GPUs that you want to run on Intel hardware.
- After installation, copy your
models,custom_nodes, andworkflowsfolders intoC:\ComfyUIif needed. - Only use custom nodes that are compatible with Intel Arc/XPU (not CUDA-only).
- If you see
Torch version: ...+xpuandDevice: xpuin the ComfyUI log, you are using Intel Arc or Intel Ultra Core iGPU acceleration.... - No need to manually patchmodel_management.pywith this setup. - To update ComfyUI or the frontend, simply run the Update batch.
- If you see
Device: cpuand notxpu:- Make sure you activated the venv before installing torch.
- Uninstall torch and reinstall the XPU version in the venv as shown above.
- If you update ComfyUI:
- Re-run the install batch if needed to ensure all dependencies are correct.
- If you get a CUDA error:
- You may have a leftover CUDA-only node or an old workflow forcing
"cuda"as device. Remove/replace these.
- You may have a leftover CUDA-only node or an old workflow forcing
- ComfyUI Official GitHub
- Intel Arc Graphics Thread (ComfyUI)
- ComfyUI install guide and benchmarks on Intel Arc (Reddit)... - How to Install and Run ComfyUI on Intel Arc - YTECHB
- Tech Craft: Install and control ComfyUI on PCs with Intel Arc GPUs (YouTube)
- ComfyUI Windows venv/conda install tutorial
MIT License. See LICENSE for details.
Scripts and guide by ai-joe-git.
Inspired by the Intel Arc and ComfyUI communities.
Share these scripts and this README to help other Intel Arc and Intel Ultra Core users enjoy a smooth ComfyUI experience!
Citations:[1] https://github.com/ai-joe-git[2] https://www.reddit.com/r/IntelArc/comments/1hhkbhs/comfyui_install_guide_and_sample_benchmarks_on/[3] https://comfyui.org/en/comfyui-windows-conda-venv[4] https://github.com/eli64s/readme-ai[5] https://github.com/comfyanonymous/ComfyUI/discussions/476...[6] https://www.youtube.com/watch?v=fQKOJVVi44E[7] https://www.ytechb.com/how-to-install-and-run-comfyui-on-intel-arc/[8] https://www.youtube.com/watch?v=iK02IBeehT8[9] https://www.youtube.com/watch?v=n2KM9ipvhaw[10] https://docs.comfy.org/get_started/introduction[11] https://benhouston3d.com/blog/crafting-readmes-for-ai[12] https://community.intel.com/t5/Intel-ARC-Graphics/Need-Step-by-step-tutorial-newbie-how-to-install-comfyui-to-run/td-p/1576043[13] https://www.youtube.com/watch?v=z5Y9L31ug4E https://www.reddit.com/r/comfyui/comments/1hhkx8l/comfyui_install_guide_and_sample_benchmarks_on/ https://www.dhiwise.com/post/how-to-write-a-readme-that-stands-out-in-best-practices https://www.reddit.com/r/LocalLLaMA/comments/1hhkb4s/comfyui_install_guide_and_sample_benchmarks_on/?tl=fr https://www.reddit.com/r/comfyui/comments/1bf7aoz/python_virtual_environments_for_comfyui/ https://docs.comfy.org/installation/system_requirements... https://www.reddit.com/r/ChatGPTCoding/comments/1hg8m52/best_practices_for_converting_documentation_to/ https://github.com/comfyanonymous/ComfyUI/blob/master/README.md https://comfyui-wiki.com/en/install/install-comfyui/install-comfyui-on-windows
Citations: [1] https://ppl-ai-file-upload.s3.amazonaws.com/web/direct-files/attachments/22300945/eb4de1d3-2fdf-4a98-9132-9b52280c51a1/paste.txt [2] https://www.reddit.com/r/IntelArc/comments/1kcp87r/guiderelease_clean_uptodate_comfyui_install_for/ [3] https://www.reddit.com/r/comfyui/comments/1kcojyl/guiderelease_clean_uptodate_comfyui_install_for/ [4] comfyanonymous/ComfyUI#476 [5] https://github.com/Stability-AI/InternalForkComfyUI/blob/master/README.md [6] https://community.intel.com/t5/Intel-ARC-Graphics/Need-Step-by-step-tutorial-newbie-how-to-install-comfyui-to-run/td-p/1576043 [7] https://cnb.cool/123123113322/comfy2222222222222222221/-/blob/master/README.md [8] https://game.intel.com/fr/stories/comfyui-vs-fooocus-for-genai-on-intel-arc-gpus/ [9] https://www.reddit.com/r/comfyui/controversial/ [10] https://docs.comfy.org/installation/desktop/macos [11] https://www.youtube.com/watch?v=fQKOJVVi44E [12] https://www.youtube.com/watch?v=iK02IBeehT8 [13] kijai/ComfyUI-HunyuanVideoWrapper#343