How to transform AUTOMATIC1111 settings to code one

General
1

I do use Colab to train my model, and I got excellent result, the issue is: I’m unable to transfer setting in my code, for example “restore faces” option I don’t see a way to have it in code who is sadly well-needed ahah

2

Depends on what options. Basically docker-diffusers-api is a wrapper around the huggingface/diffusers library. Most things supported there, we support too (but sometimes it takes a little extra work). It might be nice as a community for us to make a list of what options are called in AUTOMATIC1111 and diffusers (if there’s isn’t one already) or even docker-diffusers-api (if it’s something beyond what diffusers does but we can still offer).

Re face restoration however, it’s beyond the scope of this project* (which wraps diffusers), as those are totally different models. If you’re on Banana, I gave an example with RealESRGAN / GFPGan at GitHub - kiri-art/banana-upsample, and you can see it action at https://kiri.art/ upsample page (but it doesn’t use banana’s optimization, so loads are pretty slow - going to fix this as soon as I have time, but it’s going to be a while still).

* However, support for Stable Diffusion v2’s upsampling will be supported as soon as I have a chance.

What other options were you looking for?

3

Thank you, that help my understanding!
So if I want to achieve the same result I should create other banana deploy for this kind of work right?

what means “wraps diffusers”, it use multiple one?

4

Sure! Yeah, that’s correct.

So, what I mean by wrapping, is that, docker-diffusers-api simply takes GitHub - huggingface/diffusers: 🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch, and “wraps over it” an API in Docker.

So most of the code is just huggingface/diffusers, we put it in docker and provide an API over it… almost all the modelInputs you provide are being passed directly to huggingface/diffusers without us really touching it…so, generally speaking, anything you can do with diffusers, you can do with docker-diffusers-api (with some exceptions). Hope that makes sense!