Trained on 74 screencaps from the animation.

The dataset contains images with relatively low quality, therefore results are less detailed. Make sure to use Hires-fix for better results.

Weight 1 is fine. Don't forget to describe the character using the appropriate tags in addition to the general activation tag or negatives.

Likeliness to the original character seems to vary sometimes, could be just me though.

Just like most of my LoRas, I've trained 25 epochs, that can be requested if anyone is interested.