Connect to Automatic1111's implementation as back end?

[mikebind]

I've been successfully running stable diffusion locally using Automatic1111's webui (https://github.com/AUTOMATIC1111/stable-diffusion-webui) and wanted to try your front end (which looks cool). So, I already have the weights downloaded and have tinkered with things like textual inversion which generate additional embedding models. When I try to run UnstableFusion, I am asked for a huggingface token, which I imagine means you are going to download the model weights again. Is there a way to just point your code at the weights I already have? Or, better yet, is there a way to treat Automatic's implementation as a backend hosted on localhost? There is all kinds of innovation which has been implemented there (increasing/decreasing attention, prompt switching partway through generation, etc) which it would be great to be able to take advantage of while still using your nice innovations on the front end.

[TheSeanLavery]

Would really like this too. There is so much happening with SD, I don't want to have like 10 projects, each requiring a huge model file.

[ahrm]

Well our project uses the default SD installation that you would have if you run almost any other notebook. It is webui that is doing its own thing.

[TheSeanLavery]

I'm trying to run this without using my Hugging face info.

Is there a simple way I can do so?

I'm using some non standard trained weights of SD.

[ahrm]

You could run a your own server locally which handles /generate, /inpaint, and /reimagine post requests with the appropriate response. (See diffusionserver.py for an example). It is not going to be straightforward though.

[ndemar]

We would like to run this locally. Our internal hardware has the overhead and goes 90% idle while waiting for a queue.

[codefaux]

For the record, Windows users can create NTFS Junctions (like a symlink) to use one model file in multiple locations, using the mklink command from an elevated command prompt.

I'm using it to run for separate Stable Diffusion interfaces which don't use the official downloader, and I only use one copy of the SD model, GAN networks etc. It takes up nearly no extra space since it's a symlink.

[ZeroCool22]

For the record, Windows users can create NTFS Junctions (like a symlink) to use one model file in multiple locations, using the mklink command from an elevated command prompt.

I'm using it to run for separate Stable Diffusion interfaces which don't use the official downloader, and I only use one copy of the SD model, GAN networks etc. It takes up nearly no extra space since it's a symlink.

A very convenient Symlink Creator with GUI: https://github.com/amd989/Symlinker

Using it since a long time, works great.

[y0himba]

For the record, Windows users can create NTFS Junctions (like a symlink) to use one model file in multiple locations, using the mklink command from an elevated command prompt. I'm using it to run for separate Stable Diffusion interfaces which don't use the official downloader, and I only use one copy of the SD model, GAN networks etc. It takes up nearly no extra space since it's a symlink.

A very convenient Symlink Creator with GUI: https://github.com/amd989/Symlinker

Using it since a long time, works great.

Can you define 'Link Folder', what to name the link, then 'Destination folder' and type of link? I'd like to use the models this way since I use multiple UIs for different features. I'm trying to understand what each field means so I can create Symlinks.

Can I link to files on other physical drives? Sorry, I'm old and a newb at the same time...