I’ve been trying to get ComfyUI to work normally under windows, and sure there are some tutorials out there; mostly they’re a bit outdated or not even working anymore in my experience. After all RocM is developing “quickly” and things are changing for the better.
Instead of doing a lot of complicated stuff I browsed the github of ComfyUI and found in the latest release there was a portable version for AMD: Release v0.3.62 · comfyanonymous/ComfyUI · GitHub NICE!
I was already using the AMD adrenaline driver from the AMD website and initially didn’t have any issues, but the readme does tell you to get some newer (I guess beta) driver: AMD Software: PyTorch on Windows Preview Edition 25.20.01.14 Release Notes
I did that and all seems to work well. I managed to generate images with several models, as well as a image to 3D model and microsofts VibeVoice. All in all I didn’t run in too many issues. Sometimes the dynamic memory is not detected that well, I think by default it detects ~50GB free, so maybe for some bigger things you need to reserve it before hand in the bios.
If you’re used to the ComfyUI manager, you can install it manually: GitHub - Comfy-Org/ComfyUI-Manager: ComfyUI-Manager is an extension designed to enhance the usability of ComfyUI. It offers management functions to install, remove, disable, and enable various custom nodes of ComfyUI. Furthermore, this extension provides a hub feature and convenience functions to access a wide range of information within ComfyUI.
Because before I used the good old fooocus a lot as well, I wanted to give this a shot om my framework desktop. All I did was grab the zip file from github and replace the embedded python directory with the one that came with the ComfyUI zip. So far it seems to work well.
I hope this information is useful for people looking for solutions 
