Dreambooth repo with captioning suggestion

[victorchall]

https://github.com/kanewallmann/Dreambooth-Stable-Diffusion

I think this is a noteworthy repo that extends Xavier's Dreambooth implementation because it allows individual captions on every training and regularization image.

I've used it to create a mega model for FF7r with a bunch of characters and styles for different "districts" in the game (business, housing, slums), food trucks, and such in a single training/model.

More info here on training regime and results:

https://huggingface.co/panopstor/ff7r-stable-diffusion

or here:

https://gist.github.com/victorchall/67bc53472f86641aef1ebee1e154f5d1

[victorchall]

Inference results, all from a single model training with kane's repo:

Mixing existing SD 1.4 knowledge with newly trained stuff:
https://imgur.com/a/J8lJYrQ

And including some non-Final Fantasy examples (Tom Cruise and Emma Waton) showing the model is not like a fully-reformed video game model and can still paint "normal" stuff:
https://imgur.com/a/hVOyRmZ#6mnTC0T

[georgzoeller]

thanks I will have a look