Prior preservation

[abc123desygn]

I notice people complaining about overfitting and settings, is there any particular reason why the prior preservation flag isn't used with the dreambooth training script e.g '--with_prior_preservation --prior_loss_weight=1.0' and the notebook discourages the use of regularization when training faces?
I'm referring to this article from Huggingface https://huggingface.co/blog/dreambooth

[JustAnOkapi]

its dirty

[TheLastBen]

Early stopping the text encoder training is a better regularization method