The main problem seems to be the coding of Automatic 1111 and the faulty updates

[tabblettede]

CUDA SETUP: Problem: The main issue seems to be that the main CUDA runtime library was not detected.

[CCpt5]

It's open source . You so smart, fix it yo self.

Nothing wrong w/ the code. User error.

[tabblettede]

The user did not change anything sorry. Coders did.

[TetyLike3]

And you have evidenced the process that led up to this conclusion where?

[tabblettede]

Yes here it is, this malfunctioning came after an update. If youi have a clue thanks
The following directories listed in your path were found to be non-existent: {WindowsPath('tmp/restart')}
CUDA_SETUP: WARNING! libcudart.so not found in any environmental path. Searching in backup paths...
The following directories listed in your path were found to be non-existent: {WindowsPath('/usr/local/cuda/lib64')}

[TetyLike3]

These are the errors you received, not what you did to make up for your argument.
Have you tried using Google to look for solutions to your errors?

[tabblettede]

I did nothing particular but letting the stable-diffusion web ui do its update job, after which obviously there has been a mess in the path they provided,/ once again no personal move, all the misleading came from some coding error related to the Cuda Path.
Besides, Automatic 1111 still works after the warning with the installed version of Cuda I presume ,11.3 where it should be 11.8

[BetaDoggo]

Last update in stable was August 31st. Don't use dev branch if you can't handle things breaking.

[freecoderwaifu]

As mentioned, the UI doesn't auto update unless you do a manual git pull, and the main branch hasn't updated in a while. If you're on dev, it's expected to have some errors, most of these are fixed by deleting the venv folder and letting the launch .bat remake it, deleting the config.json files can also help.

For CUDA, there's a newer version of the CUDA Toolkit you might need to install, that an extension might be needing. An extension's update might be the actual update that might be breaking it, instead of the UI update itself. The UI on its own doesn't really need the separate CUDA Toolkit, just general CUDA support provided by the drivers, which means a GPU that supports it.

If you're using the self contained installer, it might be worth just doing a manual install by git cloning the repo, but you need to install Git and Python separately beforehand.

[tabblettede]

The git pull command is included in the ui, it checks every time I launch

Thanks for the Venv hint i will try

[tabblettede]

Additional question, according to what you said I tried to recreate the venv file but in the process of reinstalling the dependencies something went wrong, Python has quit on the "press a key" prompt. So I got back to my first venv copied folder. Everything kind of went back to normal. Except the Cuda thing again.

So in case ileave the 1 st installed Stable-Diffusion-Web entire folder alone in its hard disk as a safety copy

do you think a could perform a cleaner install on another hard disk to be sure all the components will be set properly ? Or could it alter something, somewhere, in a remote windows process that will make me unable to get back to the first install ? Thank you

[freecoderwaifu]

The git pull command shouldn't be there by default, you can see the contents of the file by default here.
https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/47a44c7e421b98ca07e92dbf88769b04c9e28f86/webui-user.bat

I recommend removing it if you don't want automatic updates.

There might be a conflict between two Python versions, you could check your Path under Environment Variables in Windows and see that it points to the standalone Python install. I also recommend Python 3.10.11. It might be better to backup all your models, extensions, etc. then do a fresh git clone on a separate folder or drive yeah, including a fresh reinstall of Python.

Uninstalling Python doesn't always entirely remove its folder though, so you might need to manually delete it before a full reinstall of it. Also doing a pip cache purge might help, since I think the uninstaller also doesn't clear it, and older dependencies cache'd there could sometimes also cause issues.

[tabblettede]

Many thanks C

[tabblettede]

Solved it as planned, I made another github clone and dit not ad the git pull.
Hopefully the last version of Cuda is installed Tks

[wfjsw]

I guess you'd better off avoiding the Dreambooth extension that install bitsandbytes which does not play well with the CUDA-bundled torch. Simply removing the extension does not fix this.

And it has absolutely nothing to do with AUTOMATIC1111. It might be related to transformers, though.