Switching the branch of the colab to main `diffusers` branch

Hey @TheLastBen,

Your fast dream booth google colab is amazing - it’s super user-friendly and needs no explanation to run it 😃 I noticed that you are using a fork of diffusers that you maintain under https://github.com/TheLastBen/diffusers

Since some weeks we also have full xformers support in huggingface/diffusers and when pip installing diffusers from main and adding this PR to your dreambooth example: https://github.com/TheLastBen/diffusers/pull/3 your google colab runs exactly as fast as it does when using your fork.

Essentially all that needs to be done in the google colab is to replace:

!pip install -q git+https://github.com/TheLastBen/diffusers with 

!pip install -q git+https://github.com/huggingface/diffusers

This way you could maybe just host your adapted train_dreambooth.py example file in this “fast-stable-diffusion” repo and then fully rely on the official main branch of diffusers instead of your branch.

I think it could make a lot of sense to shift the dependency to the main branch of diffusers so that

• We as the diffusers team can take over the maintenance burden from you
• You also can benefit from all the new features that will be implemented and quickly add them to your amazing google colab. E.g. right now your google colab would already profit from faster model loading: https://github.com/huggingface/diffusers/blob/33d7e89c42e0fe7b4a277d7a5bae12ba14828dd8/src/diffusers/modeling_utils.py#L448 as well as directly have support for the stable diffusion in-painting pipeline etc…
• Also we would be happy to add a link to this repo from our dreambooth colab for people that are looking for a “zero-code / easy-to-train” google colab: https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/README.md which should be beneficial for the community

Very curious to hear thoughts and of the community 😃

Also cc @apolinario

[TheLastBen]

I’ll try again and check closely.

[TheLastBen]

Used only xformers