Would love some more info on how the model was trained

#21
by PHUNSTER - opened

Amazing work!

I'd love to learn how to train my own model on a style and would love some guidance.

  1. What codebase did you use to train this model? TheLastBen or AUTO1111-Dreambooth extension?
  2. How many images did you use to train? Did you use captions? Were they autogenerated captions?
  3. Did you use Concept/Regularization images? If so what kind?

Any points is much appreciated!

The author of this model wrote a dreambooth style training guide: https://github.com/nitrosocke/dreambooth-training-guide

Keep in mind that this is slightly outdated, since then we've started using higher learning rates and lr-schedulers, meaning less images are needed to converge. However, the guide does explain (I believe) how this model was trained.

Sign up or log in to comment