I mean, yes, but I suspect that my environment is not the same as yours. I’m running on an RX 7900 RTX on Linux, and using the Automatic1111 frontend to Stable Diffusion. I’ll give reasonable odds that you’re running Windows. On Linux, there’s some card-specific setup, I didn’t use conda (rather, pip to set up a Python venv, which I hadn’t previously been familiar with and needed to learn a bit about), I needed driver software installed, and it took some banging around, because more people are using the software stack involved with Nvidia graphics hardware than AMD hardware. Some Radeons don’t work or require Automatic1111 to use slightly different configurations. Running on a Radeon on Windows didn’t even work until very recently (though there’s similar software that can apparently use one of the 3d graphics APIs to do something similar). I also spent some time banging around in Python testing things, switched hardware at one point, and had to spend some time tuning options to do larger images on a card that had less video memory; because this tends to push the limits of the video memory on various cards, and some cards have different hardware functionality and tricks to reduce video memory usage, there isn’t just a single set of optimizations that work on all cards. So if I tell you to do what I did, it’s probably not going to help that much. Also, I set this up some months back, which is pretty much an eternity with the rate at which the software is changing.
I don’t know if someone has put out a “one click and it’s set up” software package on Windows these days that works on various hardware or what. It may be simple, it may be a big headache, and it may depend a lot on precisely what hardware you are using.
So the answer is that if you have the right graphics hardware on the right OS, and that hardware is fairly new, then you can probably set it up. But…I can’t say “run this and it’ll definitely work”. It might be trivial! But it also might be not possible on your hardware, or might involve a lot of poking away at the guts of the thing.
Automatic1111 has setup instructions on its github repository:
Hey thanks for the detailed write up I think I’ll try to set it up this weekend. Just so you know you’re not shouting into the void, I’m running RTX 2080 on windows and I have some experience setting up venv in Python :)
The only thing I’m not familiar with is Linux if that’s required but I can probably set up a VM or dual boot, I’m overdue for learning anyway!
These are benchmarks from early in the year. They’re running an older version of Stable Diffusion on a Windows machine. A RTX 2080 isn’t blistering fast – they put it at the top of the “legacy GPU box” – but it does run it.
Note that they’re using an older Stable Diffusion version and model that’s trained on 512x512 images and the current crop of models are trained on 1024x1024 images. It’s generally preferable to generate images at the size that the model is trained at, and larger models consume more VRAM.
Video memory is a limiting factor in the size of the image that you can generate.
Look further on this subreddit, but from this, it sounds like people cap out before 1024x1x024 at 8GB.
It should be possible to use an older Stable Diffusion 1.5 model trained on 512x512 images to generate at 512x512 images, which is what was the norm until earlier this year when the larger SDXL models came out. That’s half the resolution in each dimension of the images I’ve been posting. The lower-resolution models are definitely usable, but they have more trouble with fingers and toes and…well, you can look at images generated from SD 1.5 models on civitai.com. You can then upscale the output a chunk at a time, so it’s possible to wind up with high-resolution output images.
I don’t know if it’s possible to generate lower-resolution output using an SDXL model on 8GB.
There are various SD optimizations (lowvram, half precision off the top of my head, but I haven’t been looking at them for a couple months and am our of date) that you’ll probably want to try out, since they’ll let you squeeze more out of the memory.
Sorry. I know that it’d be nice to give a definitive “this is what card X can do”, but this is all pretty bleeding edge. Two years ago, none of this was possible.
Very cool, thanks!
It’s there a how-to out there for setup? I found the conda package but haven’t looked into it yet.
I mean, yes, but I suspect that my environment is not the same as yours. I’m running on an RX 7900 RTX on Linux, and using the Automatic1111 frontend to Stable Diffusion. I’ll give reasonable odds that you’re running Windows. On Linux, there’s some card-specific setup, I didn’t use conda (rather, pip to set up a Python venv, which I hadn’t previously been familiar with and needed to learn a bit about), I needed driver software installed, and it took some banging around, because more people are using the software stack involved with Nvidia graphics hardware than AMD hardware. Some Radeons don’t work or require Automatic1111 to use slightly different configurations. Running on a Radeon on Windows didn’t even work until very recently (though there’s similar software that can apparently use one of the 3d graphics APIs to do something similar). I also spent some time banging around in Python testing things, switched hardware at one point, and had to spend some time tuning options to do larger images on a card that had less video memory; because this tends to push the limits of the video memory on various cards, and some cards have different hardware functionality and tricks to reduce video memory usage, there isn’t just a single set of optimizations that work on all cards. So if I tell you to do what I did, it’s probably not going to help that much. Also, I set this up some months back, which is pretty much an eternity with the rate at which the software is changing.
I don’t know if someone has put out a “one click and it’s set up” software package on Windows these days that works on various hardware or what. It may be simple, it may be a big headache, and it may depend a lot on precisely what hardware you are using.
So the answer is that if you have the right graphics hardware on the right OS, and that hardware is fairly new, then you can probably set it up. But…I can’t say “run this and it’ll definitely work”. It might be trivial! But it also might be not possible on your hardware, or might involve a lot of poking away at the guts of the thing.
Automatic1111 has setup instructions on its github repository:
https://github.com/AUTOMATIC1111/stable-diffusion-webui
There’s also a Stable Diffusion subreddit – no community here the current level of activity yet:
https://old.reddit.com/r/StableDiffusion/
There’s information that you can find there about specific problems.
If you want to use a nonstandard model – which I do – then many people have uploaded them to civitai.com, and they can be found there.
Hey thanks for the detailed write up I think I’ll try to set it up this weekend. Just so you know you’re not shouting into the void, I’m running RTX 2080 on windows and I have some experience setting up venv in Python :)
The only thing I’m not familiar with is Linux if that’s required but I can probably set up a VM or dual boot, I’m overdue for learning anyway!
Linux is not required, certainly not if you’re running Nvidia hardware.
VRAM is going to be very tight. It looks like an RTX 2080 has 8GB. Stable Diffusion is very demanding of VRAM.
https://www.tomshardware.com/news/stable-diffusion-gpu-benchmarks
These are benchmarks from early in the year. They’re running an older version of Stable Diffusion on a Windows machine. A RTX 2080 isn’t blistering fast – they put it at the top of the “legacy GPU box” – but it does run it.
Note that they’re using an older Stable Diffusion version and model that’s trained on 512x512 images and the current crop of models are trained on 1024x1024 images. It’s generally preferable to generate images at the size that the model is trained at, and larger models consume more VRAM.
Video memory is a limiting factor in the size of the image that you can generate.
googles
https://www.reddit.com/r/StableDiffusion/comments/wtd4e1/if_i_want_to_generate_large_images_say_1024x1024/
Look further on this subreddit, but from this, it sounds like people cap out before 1024x1x024 at 8GB.
It should be possible to use an older Stable Diffusion 1.5 model trained on 512x512 images to generate at 512x512 images, which is what was the norm until earlier this year when the larger SDXL models came out. That’s half the resolution in each dimension of the images I’ve been posting. The lower-resolution models are definitely usable, but they have more trouble with fingers and toes and…well, you can look at images generated from SD 1.5 models on civitai.com. You can then upscale the output a chunk at a time, so it’s possible to wind up with high-resolution output images.
I don’t know if it’s possible to generate lower-resolution output using an SDXL model on 8GB.
There are various SD optimizations (lowvram, half precision off the top of my head, but I haven’t been looking at them for a couple months and am our of date) that you’ll probably want to try out, since they’ll let you squeeze more out of the memory.
Sorry. I know that it’d be nice to give a definitive “this is what card X can do”, but this is all pretty bleeding edge. Two years ago, none of this was possible.