I still can't quite get the hang of training a LoRA.

The result comes close to what I want, but once again I apparently messed up the parameters, so I have to sift through a ton of variations before something finally satisfies me.

For now, I'll leave it as is — with its help, I'll generate a bunch of images for the dataset.

Tags:
1man, angry, back, bare, black_cape, black_eye, blue hair, bob, chest, cloak, collar, darcia_wr, blue eye, eyepatch, feather_headdress, forehead_jewel, greyscale, heterochromia, high, high_collar, hood, long, long_cape, monochrome, portrait, purple_cape, red_smile, shirt, short, style_patch, volto_mask, warped_mask

________________________________________________________________________________________

Oh, and another problem popped up here — I usually draw anime girls, so my checkpoints are all tuned for that style, and not all of them can draw a Volto mask properly. Out of 26 checkpoints, only 3 managed to handle it.

And now I'm wondering — can the base model I trained on even draw it???