You must log in or # to comment.
Cool project!
Would love AMD Linux support though, Stable Diffusion is fairly easy to install, but i just don’t like the Stable Diffusion WebUI
A Flatpak would be golden
Just install ComfyUI locally, use their AMD Linux install guide, then install the Models and Extensions that KritaAI needs, as mentioned on their github. This way you can easily use the full Comfy, which is what KritaAI installs an the backend, and just select local Server in the KritaAI options to usw your local install. Works perfectly well on my Arch system with a 6800XT.
Aw nice to see some AI initiatives making it into free software too and not only closed ones.
Most AI are OpenSource, but mostly paid when they are as Online service, server cost money, also with OpenSource.
I meant the integration, I only heard about Photoshop having a integration but not Gimp for example.
Giving it a try now! Haven’t tried Krita before either.
Edit: the server seems to error out when I try to run it, testing with the CPU version instead. Just need to wait for it to finish
This is very cool, sadly AMD GPUs in Linux are not supported. I hope the server component can be run independently there at some point.
I’ve got an RX 580 8GB. What sucks is that it USED TO SUPPORT running AI and stuff. But AMD removed that support in new versions of the driver. Might have had a good reason, might not have. Still, sucks.
it USED TO SUPPORT running AI and stuff. But AMD removed that support in new versions of the driver.
Can you run old versions? Which driver are you talking about?
My memory sucks, but I’ll try to find an example for you. I remember it apparently last worked like two major Ubuntu releases ago, 21.04 or something. I tried for weeks to find workarounds and community-updates to fix it. I was on Pop!_OS at the time. Nothing quite worked and I gave up.
https://www.reddit.com/r/StableDiffusion/comments/ww436j/howto_stable_diffusion_on_an_amd_gpu/
Am I understanding correctly and this is truly FOSS and fully offline, there’s no remote server or model we have to connect to? What was the model trained on? I’m really curious but I also don’t want to support proprietary unethical data sourcing.
It create a local server, in the setup (~10GB disk), as say, selfhosted.
What would be the minimum requirements outside of the disk space?
I assume I’m hitting a limit, but I’m not sure what kind of upgrade I’d need to use it effectively
I remember reading something on GitHub. I think it should work on a current PC/Laptop, the question is what resolution you can render and how long it will last.
