Feature Request: conditional dropout

[forestsource]

Hi! Thank you for your great works.

It would be nice if you could support for conditional dropout. https://github.com/victorchall/EveryDream2trainer/blob/c816e2577379060cd1015b6a1bca5a109049ab6a/doc/ATWEAKING.md#conditional-dropout https://github.com/victorchall/EveryDream2trainer/blob/98f9a7302d3325354e69bfa87f3851e98c4735e9/data/every_dream.py#L145-L156

[kohya-ss]

It looks good!

The implementation drops a whole prompt (change to the empty string), but it may be also good to drop one of tags.

[lolxdmainkaisemaanlu]

I think Stable Diffusion 1.5 was finetuned with 10% conditional dropout. That lets it work across a wide range of Classifier Free Guidance (CFG) values.

I think this implementation would be very helpful for training and give better results.

[tkgix]

ドロップアウトによって結果が非常に大きく異なる方です。 設定した確率に応じてキャプションとタグを空で学習させる単純なコードなので、早く適用されるといいと思います。