Loading huggingface textual inversions

[tmm1]

🤗/sd-concepts-library has prettained textual inversions: https://huggingface.co/sd-concepts-library

would be nice to be able to easily load and use these

cc AUTOMATIC1111/stable-diffusion-webui#151

[lstein]

The ability to read hugging face .bin files was just added today. You'll find it in the development branch, and documented in docs/features/TEXTUAL_INVERSION.md

[tmm1]

Thank you!

#498

https://github.com/lstein/stable-diffusion/blob/development/docs/features/TEXTUAL_INVERSION.md

[tmm1]

Looks like training doesn't yet work on M1. Has anyone started looking into this already?

EDIT: #517 (comment)

[tmm1]

I confirmed that this does work.

For example, download https://huggingface.co/sd-concepts-library/ugly-sonic/blob/main/learned_embeds.bin and save as uglysonic.bin

Then run scripts/dream.py --embedding_path uglysonic.bin and you will see:

>> Model loaded in 13.71s
Added terms: *, <ugly-sonic>

Then use the prompt "a photo of <ugly-sonic>" and you will get a photo: