Spaces:

Duplicated from  anzorq/sd-to-diffusers

diffusers

/

sd-to-diffusers

like 71

Running

Any tips to make it working again ?

yeah, anyone has a workaround?

The same problem for me, both HuggingFace space and GoogleColab are not working

Works for me. Is it giving any type of error?

It produces empty pull requests like these:
https://huggingface.co/NickKolok/meryl-stryfe-20230123-2200-rep-4800-steps_1/discussions/6
https://huggingface.co/NickKolok/meryl-stryfe-20230123-2200-rep-4800-steps_1/discussions/5
etc.

I've tried both for usual .ckpt (DreamBooth-trained) and -inpainting.ckpt (merged via AUTOMATIC1111 with this guide: https://old.reddit.com/r/sdforall/comments/zyieht/how_to_turn_any_model_into_an_inpainting_model/)

No error is given, but the generated pullrequest is empty.

Same here. Opened PRs are now empty.

We did a factory reboot on the space and it works now for me. Can you please try again?

We did a factory reboot on the space and it works now for me. Can you please try again?

I've just tried with inpainting model, still no changes:
https://huggingface.co/NickKolok/meryl-stryfe-20230123-2200-rep-4800-steps_1/commit/756fd49c084b5a1a5271af74833f6ab5c4a56932

I've tried both EMA and non-EMA options, v1 structure.

The file is: https://huggingface.co/NickKolok/meryl-stryfe-20230123-2200-rep-4800-steps_1/blob/main/meryl-stryfe-20230123-2200-rep-4800-steps-inpainting.ckpt (the links does not lead to a raw file but to a page on HF)

Anyway, you are doing a really great and important work! If we have some (three?) formats, the whole industry needs a solid tool to convert the data from any format to another!

I've tried non-inpainting model, and it looks working! https://huggingface.co/NickKolok/meryl-stryfe-20230123-2200-rep-4800-steps_1/discussions/13
The converted file was: https://huggingface.co/NickKolok/meryl-stryfe-20230123-2200-rep-4800-steps_1/blob/main/model.ckpt (again, not raw link)

So, we can suppose that the problem may be stuck to:
a) AUTOMATIC1111-produced merges;
b) inpainting models.

Do the latter have different structure? Can I use a .ckpt file in Python with diffusers in a manner I use for usual, "multi-file" models?

I did not test the in-painting models, only the "normal" ones, so it's good to have confirmation that they work now :)

The in-painting models have a slightly different architecture (more channels as input to the UNet), so they wouldn't work with the normal yaml. I'll check if they can be converted with this tool.

Hi,
I searched a lot for a solution for this and were super happy to find this space, this tool will be very useful. Unfortunately I also seem to end up with empty PR's. I am trying with a ckpt model that were generated by training a SD 1.5 model using Dreambooth extension in Automatic 1111's interface. Any ideas on how to get it working? Thanks!

Same. Empty PRs right now.