Recommendations:

Lora Weight: 0.3-0.7

CLIP SKIP 0-1

DivineEleganceMix or AbyssOrangeMix3 | models

The resolution of the original dataset images leaves much to be desired, so I recommend you to use upscale ( or HighRes fix ) on your images.

Additional train info:

Trained on: AbyssOrangeMix3.

Trained on 20+ anime images.