>>100313015Is having 1116 images to train artist's style LyCORIS LoRA overkill?
It's an artist with engraving and cross-hatching style artworks.
I was training this LyCORIS with only 200 images but the result wasn't quite satisfying.
It could only copy the engravings/hatches decently when I prompt the same thing as the captions in the dataset and when I strayed away from it, it gave me hues instead.
How do you train artist's style LoRA to retain their intricate details while still being flexible with the prompts?
Do I lower the batch size?
Do I increase the net dim/alpha?
Do I increase the LR?