[WIP] Latnet (and Stable) diffusion text2img conversion script.

[patil-suraj]

This PR adds conversion script for text2img latent diffusion models from CompVis.

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[kabachuha]

@patil-suraj Do you have any plans for reverse functionality? I.e. diffusers .bin -> original .ckpt? Your knowledge would help us greatly

[pcuenca]

Hi @kabachuha! Work towards that goal is being coordinated in issue #672, where you have participated :) It looks like @rashmimarganiatgithub is taking a go at it!

[NicholasKao1029]

Are there docs for this?

[patil-suraj]

Hi @NicholasKao1029 you can check the docstring here, the script can now be called as a function https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/convert_from_ckpt.py#L778

[NicholasKao1029]

Thanks @patil-suraj is it correct to say that this conversion is a perfect 1:1.

As in ckpt 1 -> diffusers -> ckpt 2 ckpt 1 == ckpt2

[patil-suraj]

Yes, it should be 1:1

[NicholasKao1029]

Very nice thanks, can I also confirm that safetensors transformation is the same way (also outputting to a safetensors)

[patil-suraj]

The script convert ckpt safetensors to diffusers but with native torch doesn't output safetensors yet