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.