Man, this character has to be the hardest one to train ever or at least its pretty close to that. You only realize the complexity and contrasts of the outfits when you are making the datasets.
TLDR: Its a FAIL. Lots of problems. I won't describe them otherwise I'll just end up writing a wall of text and you don't want that. You will notice when you try it but most of them are solvable with helper tags + random seeding.
Recommended tags combination (trained triggers + not trained helpers):
yuna \(kuma kuma kuma bear\), red eyes
black bear costume, black bear hood
white bear costume, white bear hood
hood down
bear puppet gloves, black bear puppet with blue ribbon on right hand, white bear puppet with pink ribbon on left hand
bear footwear, black bear footwear on right foot, white bear footwear on left foot
black and white gothic dress, black ribbon, long white skirt, choker, puffy short sleeves, white flower with black ribbon on hair
Description
first and likely last release.
trained at LR=0.00003 with ~160 samples: ~130 screenshots + ~10 fan art + ~20 AI generated images with an earlier checkpoint and Flux Klein assistance to mitigate the fact the character is pretty much ALWAYS wearing the entire outfit. By making a few AI samples with just gloves/footwear/main outfits I was able to make this LoRA more versatile. You can even use it just to draw specifics parts of the outfits and nothing else.