Fine-tuning #StableDiffusion2 768x768 with #DreamBooth, text-inversion & vanilla fine-tuning is now supported out of the box in 🧨 diffusers. Just choose the right checkpoint & resolution & you are good to go!
Github:github.com/huggingface/di
Colab links below⬇️
Some early results
Conversation
Replying to
2
4
33
The scripts and colab now support fast and memory_efficient_attention by default, just install xformers, no need to fork the library anymore!
1
1
16
Note that, for dreambooth when text encoder fine-tuning is enabled it does not (yet!) fit in 16GB VRAM because of the high-resolution and big text-encoder. But pretty sure the community will figure out a way to make it work. Stay tuned!
2
2
19
Awesome thank you so much for the update!
Now we just need a update for the ckpt conversion script please ✨
5
This is more of a question for AUTOMATIC but have you seen v2 finetuned chkpt work on the Automatic GUI?
6