Discuss matters related to our favourite AI Art generation technology
It seems like you are missing a volume mapping from your PC to your Docker. The way it works is that you would put the inside the folder on your PC and it would automatically appear in docker. Could you provide the command you are using? Or if you are using docker-compose the docker-compose.yml?
I used the command to install DS found on the git page
docker compose --profile download up --buil
And then to run Automatic1111 I used the following command
docker compose --profile auto up --build
You should put your models inside the folder model/StableDiffusion that is in your host machine.
Actually I tried running it, the correct folder would be. [PATH_TO_WHERE_YOU_GIT_CLONE]/stable-diffusion-webui-docker/data/models/Stable-diffusion
So I had a sudden thought, went back and tried a different model file that is a .ckpt file rather than the .pth files I was testing with. This worked, so obviously it was a simple matter of the wrong file type. Now I feel foolish!
Out of curiosity, what is the .pth file type used for?
So far I only have seen .pth in upscalers. Inside the folders ERSGAN and RealERSGAN for example. And yes models are .chkpt or .safetensors
I see. Well thanks for your help anyway mate 🙂
Never install .ckpt checkpoints (unless you're 1000% sure of the source. Always use safetensors files instead.
Hmm, that's a good point. I've just looked up the difference. Thanks
I've tried that. There is already a folder at /data/models/Stable-diffusion that has two .ckpt files that match the options that are available in the web ui. I've tried pasting the other .pth files that are in other folders within /data/models but they do not appear as options in the ui.
Just to be thorough, does your docker run command / docker-compose have a volume field or flag to link your folder from your host to the container? Or are you making these folders from within the container?
Forgive me for not knowing the details here, I've never used Docker before so I'm just following the guide.
I first created a folder, then I cloned the git repo to it. Finally, from the command prompt, I ran the install command from within the folder.
[docker compose --profile download up --build]
Show me your docker-compose.yml contents. Won't be able to reply until later in the morning. But I should be able to help you get set up.
So it turns out I was using the incorrect file type. I had chosen a set of model files that came with the install that were .pth files. These didn't work, but when I tried it with .ckpt files it works as expected. Big sigh. Thanks for the help anyway mate!
Nice! Glad you got it figured out
I've never used the docker version, it's been easier to use the installer zip.
Have you tried refreshing the list of models?
I did after each attempt.
Using the installer zip, can you just drop model files into the correct folder and have them work? If so, I may have to just use that version instead.
Yes, after it's installed you can just shove the safetensors etc into the folder and they show in the list after refresh.
Thanks. The Docker version is supposed to be easier, as it handles updates to all dependencies automatically. But, if I can't add new models there's not much incentive to keep using it.
I'm sure there is a way you can, but I have no idea, I don't really understand docker. I never use it
Searching through the issues, this seemed relevant:
there are a couple of complicating factors.. One is that all the different modules (stable diffusion, esrgan, realesrgan, clip, blip, codeformer, etc..) have their own functions to download models and their own preferred places to save them, and currently not all of them are configurable. Would be nice if they could be changed so they look in models// so just the models dir could be mapped to the host file system and kept outside the image.
So it reads like it’s going to be dependent on the module and all of the module’s specific nuances.
module or model? Because for models it's blindingly easy.
I'm trying to add different models, not modules.
Arg, I didn't spot that. That's frustrating. Thanks