AMD for ComfyUI - How to Install Stable Diffusion with Radeon GPUs and APUs on Windows

Learn this advanced method for using ComfyUI with AMD hardware - this should be fine with the latest version of ComfyUI (Jun 2024) and using the APUs and discreet GPUs.

Beginner's Guide to ComfyUI and Stable Diffusion

As an Amazon Associate Pixovert earns from qualifying purchases, as well as from associations and partnerships with other retailers, advertisers and sponsors.

Recommended GPUs

APUs are fine, but performance will be limited.

Detailed instructions

Join 🏆 this channel to get access to exclusive content and perks:

[Pixovert]

Using ComfyUI with AMD hardware - this should be fine with the latest version of ComfyUI (Jun 2024) and using the APUs and discreet GPUs.

Beginner's Guide to ComfyUI and Stable Diffusion

As an Amazon Associate Pixovert earns from qualifying purchases, as well as from associations and partnerships with other retailers, advertisers and sponsors.

Recommended GPUs

APUs are fine, but performance will be limited.

Detailed instructions


Join 🏆 this channel to get access to exclusive content and perks:

[enderepsilon]

You're my favorite person. webui has not been working but this just worked first try. Thank you so much

[roki]

I just use latest version of Amuse, its simple and elegant, offers enough good models and options to have fun. With some more optimisations and new models it will become great option less demanding users on AMD. Works well on 7800xt. I also have automatic1111 and comfy installed but i cant force my self to use them...

[YouArentHereYet]

Fyi in the requirements.txt file I had to change "torchaudio" to "torchaudio>=2.3.1" because it was trying to install version 2.4.0 which is not compatible with torch 2.3.1, which was what was installed by default.

[Ghadrat]

The tutorial works, I manage to install it but I can only create one image and then I have to close and open again because it shows an error of low available memory, I have a 12GB RX 6750 XT, Ryzen 5 5600 and 16GB of ram

[RenHQ.]

Help! I was able to make it work but then I closed it when I was done. How to I activate it again using the Anaconda prompt?

[andresbares]

Nice tutorial, thanks!
Is there a faster way to launch comfyui than trough the anaconda console next time I run it?

[__-fixg]

you are running it on the integrated gpu of your cpu as far as i can tell since its marked as gpuTM instead of its actual model number

[beragis]

I have had trouble with directml not freeing memory and eventually causing my PC to hang, thankfully I found a fork of ComfyUI that uses ZLuda, you might want to try that fork to see how it compares.

[bubbabenali]

Thank you for this tutorial, but man openML is so slow... On a 1440x640px picture my 7800xt only gets to ~2, 5sec/iteration while my 4060ti/16gb gets 2-4 iterations/sec.
I just hope there will be some usefull zluda all in one solutions soon. All the current ones won´t survive the first reboot for some reason.