Request for Stable Diffusion WebUI Adaptation for Windows on ARM with Snapdragon X Elite Processor

[Ryanwu316]

Hello Stable Diffusion WebUI Developers,

I hope this message finds you well. I am using a Windows on ARM (WoA) computer equipped with the Snapdragon X Elite processor, which is the first AI PC on the market. This processor boasts a powerful 45 TOPS NPU, providing significant AI capabilities.

However, I have encountered compatibility issues when trying to run the Stable Diffusion WebUI on this setup. Given the unique architecture and the AI acceleration features of the Snapdragon X Elite, I believe there is a significant opportunity to optimize and adapt the Stable Diffusion WebUI for this platform.

1. Are there any current plans to support Windows on ARM devices, specifically those with the Snapdragon X Elite processor?
2. What steps would be necessary to begin adapting the WebUI for this architecture?
3. Are there any existing projects or branches that are working on compatibility with ARM-based Windows systems?
4. How can the community contribute to or support this development effort?

I appreciate any guidance or resources you can provide. Making the Stable Diffusion WebUI compatible with Windows on ARM would greatly benefit users like me who are leveraging this new technology.

Thank you for your time and consideration.

[nkedel]

As far as I know, there's no PyTorch version for either Snapdragon X NPU or GPU, so you're limited to CPU processing.

There ARE CPU-only pytorch builds for Windows ARM64: https://blogs.windows.com/windowsdeveloper/2025/04/23/pytorch-arm-native-builds-now-available-for-windows/

You can try the following steps:

1. install python via the windows store (I already had 3.13 which is a bit new for automatic
2. run webui.bat in a cmd window
3. it will fail to install safetensors, but it creates the venv for you
4. install torch and torchvision via the URL in the link above
5. pip3 install -r requirements.txt -> some thing will install, and then safetensors will fail to install (I guess no native arm version?
6. install rust -> https://rust-lang.org/tools/install/
7. close the cmd window
8. reopen the cmd window and activate the venv
9. iterate running pip3 install -r requirements.txt (and possibly patching the requirements.txt versions) until you get it to install cleanly

I get stuck on scikit-image

Will update this if I find a way to get past that