I've kept training the model until step 750, where it seems close to overtraining already. At the same time, it also looks more accurate than the previous version and also offers better results on a consistent basis, at least with the prompts and models I've been testing (mainly StablyDiffused and DreamShaper).
I've kept training the model until step 750, where it seems close to overtraining already. At the same time, it also looks more accurate than the previous version and also offers better results on a consistent basis, at least with the prompts and models I've been testing (mainly StablyDiffused and DreamShaper).