English | 中文说明

ComfyUI Windows Portable Package for Intel GPU|XPU

Similar to the official ComfyUI Standalone Portable package, but built for Intel GPUs.

This project is derived from ComfyUI-Windows-Portable.

Supported Intel GPUs (CPUs)

• Arc B

  • B580, B570, and subsequent Battlemage series GPUs

• Arc A

  • A770, A750, A580, A770M, A730M, A550M, Flex 170

• Arrow Lake-H

  • 285H, 265H, 255H, 235H, 225H

• Lunar Lake-V

  • 288V, 268V, 266V, 258V, 256V, 238V, 236V, 228V, 226V

• Meteor Lake-H

  • 185H, 165H, 155H, 135H, 125H, 165HL, 155HL, 135HL, 125HL

For the full list, refer to Intel Documentation.

Other Features

• Only a few functional models are pre-installed. Users please prepare SD models of your favorite.

• Pre-installed 40+ commonly used custom nodes. See Full List.

  • Dependencies (200+ compatible Python packages) are pre-configured, including those that require compilation like insightface and dlib.

  • Pre-downloaded some model files, including those for ReActor Node and Impact Pack, to avoid download delays during the first launch (which may cause freezing for users with a poor Internet connection).

• Built using GitHub CI/CD pipelines for easy DIY. Users don’t need to configure CI/CD; just fork the repository on GitHub and run the build script. See Creating Your Own AIO Package.

How to Use

1. Install the latest Intel GPU Drivers.

2. Install the latest VC++ Runtime ( Microsoft Page | Direct Download )

3. Download the package from the Releases Page, or use the direct download links:

   • Arrow Lake-H (Ultra Series 2 Mobile Processors) users:

     • ComfyUI_Windows_portable_arl.7z.001

     • models.zip.001 (Model package, only needs to be downloaded once, no need to re-download for future updates)

   • All Other Intel GPUs users:

     • ComfyUI_Windows_portable_xpu.7z.001

     • models.zip.001 (Model package, only needs to be downloaded once, no need to re-download for future updates)

4. Extract the .7z.001 file (recommended not to place it in a deep directory), and extract the .zip.001 file to the same directory.

5. Place SD models under ComfyUI\models\checkpoints.

   • You can create subdirectories for better organization.

6. Run RUN_Launcher.bat to start.

• After launching, the program will automatically open a browser. You can also manually access: http://localhost:8188/.

• Close the window to exit the program.

Known Issues

Test results on Arc B580 (12G) are as follows:

SD 1.5	🆗 Fast
SDXL	🆗 Fast
Playground 2.5	🆗 Fast
Stable Cascade	🆗 Very fast
FLUX.1	❌ Crashes during the first step of loading the model, no error message, just exits
SD 3.5	❌ Crashes during the first step of loading the model, no error message, just exits

Current known issues:

1. When VRAM is insufficient, the driver does not actively use system memory but crashes directly, and sometimes does not release VRAM after crashing.

   • If the program crashes during the sampling phase, try reducing the latent image size.

   • If the program crashes during model loading, there is no good solution for now; try switching models.

   • Closing GPU-accelerated programs (e.g., browsers) can free up some VRAM.

2. Switching model types (e.g., running an SDXL workflow after a Stable Cascade workflow) may cause the program to crash. Adding the --disable-smart-memory parameter by default alleviates this issue.

   • Using "Disable Smart Memory Management" will increase model loading time. If you only use a single model after launching, you can disable this option to save time (but be cautious of VRAM overflow).

3. The driver performance on Windows differs from that on Linux (kernel module xe), mainly in VRAM usage, with little difference in inference speed. WSL2 has not been tested.

Usage Tips

1. The ExtraScripts folder contains command-line launch scripts, which are equivalent to the launcher. If you are a user of ComfyUI-Windows-Portable, you should be familiar with this.

2. Make good use of ComfyUI-Manager (the Manager button in the top-right corner of the ComfyUI page) to manage custom nodes: install, update, disable, and uninstall.

   • Be cautious with "Update All"; there’s no need to update unused nodes to avoid Python package conflicts.

3. If you dislike temporary files and download caches generated by the program, consider using Sandboxie to constrain them and isolate the environment.

   • When using Sandboxie, for I/O performance, it is recommended to configure the program’s main directory (ComfyUI_Windows_portable) as "Open Access" in "Sandbox Options" → "Resource Access".

4. External Model Directory

If you place model files on different partitions or share a set of model files across multiple ComfyUI instances, you can configure ComfyUI to load external model directories:

• Rename extra_model_paths.yaml.example in the ComfyUI directory, removing the .example suffix.

• Edit extra_model_paths.yaml, where lines starting with # are comments.

Reference file (click to expand)

comfyui:
    base_path: D:\models\
    animatediff_models: animatediff_models
    animatediff_motion_lora: animatediff_motion_lora
    bert-base-uncased: bert-base-uncased
    checkpoints: checkpoints
    clip: clip
    clip_vision: clip_vision
    configs: configs
    controlnet: controlnet
    depthfm: depthfm
    diffusers: diffusers
    diffusion_models: |
        diffusion_models
        unet
    embeddings: embeddings
    facerestore_models: facerestore_models
    gligen: gligen
    grounding-dino: grounding-dino
    hypernetworks: hypernetworks
    insightface: insightface
    instantid: instantid
    ipadapter: ipadapter
    loras: loras
    mmdets: mmdets
    onnx: onnx
    photomaker: photomaker
    reactor: reactor
    rembg: rembg
    sams: sams
    style_models: style_models
    text_encoders: text_encoders
    ultralytics: ultralytics
    unet: unet
    upscale_models: upscale_models
    vae: vae
    vae_approx: vae_approx

5. More External Model Directories

ComfyUI has five common locations for saving model files:

• The built-in ComfyUI\models directory.

• External model directories configured via extra_model_paths.yaml.

• Model files downloaded via HuggingFace Hub (HF official downloader).

• Model files downloaded via PyTorch.

• Files downloaded in-place by nodes in ComfyUI\custom_nodes.

Among these:

• HF Hub defaults to downloading files to C:\Users\UserName\.cache\huggingface\hub.

• PyTorch defaults to downloading files to C:\Users\UserName\.cache\torch\hub.

This package modifies the launch script to redirect these to the program’s root directory, under HuggingFaceHub and TorchHome folders, respectively, for easier management. If needed, you can edit the launch script to change these paths.

6. How to Perform a "Refresh" Major Update

This method does not use ComfyUI-Manager for updates but directly replaces the package with a new version. If the process goes smoothly, it avoids installing or upgrading Python packages, thus preventing dependency conflicts.

1. Extract the new package.

2. Delete the ComfyUI, HuggingFaceHub, and TorchHome folders in the new package.

3. Copy (or move) these three folders from the old package to the new one.

4. Run RUN_Force_Updater.bat in the new package’s root directory.

5. If any nodes fail to load, use ComfyUI-Manager to "try fix."

Pre-installed Custom Nodes List

Workspace
ComfyUI Manager	Crystools	ComfyUI-to-Python-Extension
General
AKatz Nodes	Comfyroll Studio	ComfyUI Essentials by cubiq
Derfuu Modded Nodes	Custom Scripts by pythongosssss	Efficiency Nodes by jags111
Jovimetrix	KJNodes	Mikey Nodes
Mira Nodes	rgthree Nodes	smZ(shiimizu) Nodes
Use Everywhere	WAS Node Suite	ComfyUI-Easy-Use
Control
Advanced ControlNet	ControlNet Auxiliary Preprocessors	Detail Daemon
IC-Light Native	Impact Pack	Impact Subpack
Inspire Pack	InstantID by cubiq	IPAdapter plus
Layer Style	PuLID by cubiq	LayerDiffuse
Portrait Master	ReActor Node	SD Dynamic Thresholding
SDXL Prompt Styler
Video
AnimateDiff Evolved	FizzNodes	Frame Interpolation (VFI)
MTB Nodes	Video Helper Suite
More
ComfyUI-GGUF	Depth Anything V2 by kijai	DepthCrafter by akatz
Face Analysis by cubiq	Florence-2 by kijai	Image Selector
ProPost	Segment Anything 2 by neverbiasu	Ultimate SD Upscale
WD 1.4 Tagger

If you encounter compatibility issues after installing new nodes, you can disable conflicting nodes in ComfyUI-Manager.

Creating Your Own All-In-One Package

This repository utilizes a pipeline to build the package, and the codebase doesn’t contain specific configurations or require additional access permissions. Hence, you can directly fork this repository to start executing the GitHub Workflow.

1. After forking, find Actions on the page.

2. Locate Build & Upload Package.

   • For example, the page in my repository looks like this.

3. Click Run Workflow.

4. Wait about 20~40 minutes until the workflow run complete.

5. Go to the releases page of your repository, where you will find the newly generated draft, ready for download or editing.