Thread 106005551

I've been trying to get stable diffusion to work on linux for 5 months now

>>106005551 (OP)
Quite the resilient spirit you have. Even in the face of abject idiocy, you continued for 5 months.

>>106005551 (OP)
Clean my neetcave instead. It will require similar perseverance but in the end you will accomplish something.

>>106005551 (OP)
holy crap anon... just use stability matrix and use reforge webui on it

>>106005551 (OP)
Install Easy Diffusion, retard.
IF you have AMD, then I can feel your frustration, but even those shit GPUs can run stable diffusion now.

>>106005633
It's giving me more issue than the more complicated frontends. Won't detect GPU (terminal confirms I have the right drivers) and won't even let me use models other than the one it came with.

>>106007140
What is your GPU? If it is NVIDIA it should detect it automatically.
The project will create this folder where you can dump your models
Easy-Diffusion-Linux/models/stable-diffusion/
Same for lora files
Easy-Diffusion-Linux/models/lora/
And custom vae files
Easy-Diffusion-Linux/models/vae/

>>106007155
I use a nividia rtx. I even tried different nividia drivers from the driver manager and tried a terminal installation, but it still won't let me switch gpu usage on. The model path goes: easy-diffusion/models/stable-diffusion. The model shows up but won' let you gen anything due to the "local_files_only" error which I don't understand.

>>106005551 (OP)
I've been thinking of trying to get stable diffusion to work on linux for 5 months now, but kept putting off to maybe later

I'm glad to see my lack of hard work paid off as much as your hard work.

got it to work with rocm but it constantly OOM at various point. Often i can't even load a model. I stick to zluda on windows even though it's half the speed.

As per usual you gotta wait 5-10 years then maybe Linux will support what Windows and Mac currently supports

nvidia + linux + AI = Bad Time

shoulda got AMD to do things 25x slower but hey it'll work

>>106005551 (OP)
>create a venv
>install ROCm, python and SD in it
>run
I have a 7800 XT which gains first party driver support recently and it just werks

>>106007853
Yeah venv seems to be key
Less Python dependency hell that way (but still some because it's still Python crap)

Just wished they used a proper programming language to build all the AI tools in.

>>106005551 (OP)
based

>>106007934
Just use docker

>>106005551 (OP)

distrohopping again not even 480p is smooth at youtube

I'm running it on debian stable, installed nvidia drivers and cuda from repo. Then just git clone. It was still from automatic1111 when I first installed it.
Started out on a 3060, but swapped it to 4060TI for SD1.5 and later XL no need to update drivers

>>106007934
Just use uv you nigger

>>106005551 (OP)
Should've installed Gentoo

>>106005551 (OP)
>>106005633
Is it actually that hard to run with AMD on linux? I assumed it would be pretty similar, maybe with the added issue of installing certain libraries.

>>106011391
Just looking into it it looks like there's some extra steps for AMD, but not really anything unexpected on linux. At least for AUTOMATIC1111.

>>106011422
Even further it seems like so long as you're on Ubuntu and don't have an old ass AMD driver it shouldn't be too difficult except maybe some CUDA check issues on startup.

For me is extremely simple. NVIDIA is the easiest bet and work flawlessly. AMD used to be unusable, took them long enough, now you have to always do tweaks but better than nothing I guess.

>>106011560
How did you get it to recognize your nividia card?

>>106007835
>nvidia + linux + AI = Bad Time
werks just fine on my machine. in fact my nvidia card just werks with everything on my machine. hmmmmm

>>106011723
Do you people just never read the documentation?

>>106011739
same here

>>106007232
Did you install CUDA drivers?

>>106011723
It was detected automatically. I'm on Fedora though.

>>106011765
Reading is for trannies.

I got it to work on Arch Linux with my RTX 5080 in one afternoon last month. The trick was installing it in a Docker container instead of torturing myself by trying to use system libraries.