Recommended settings
LoRA network weight 0.6-0.9
Please generate using highres fix for best effect
For LoRA Block Weights:
For her default appearance, dropping OUTS blocks (1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0) seems good.
For doing other outfits, or getting more freedom of poses/composition, using only OUTD blocks (1,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0) gives more flexibility, but her age appearance will need to be prompted for
Suggested prompts
Regular Outfit (S3 anime) (Pre-vol 15)
1girl, closed eyes, white hair, two-tone hair, long hair, tiger ears, tiger tail, white dress, multicolored dress, green dress, short sleeves, cross-laced, black skirt, twin braids, low twin braids, hair ornament, bracelet, black footwear
This LoRA was primarily trained on this outfit
Known issues:
Her hair ornaments and bracelets sometimes double up. Sometimes this can be fixed on highres fix using Euler a
The model will only sometimes put her tail below her skirt, and can often depict it too long. SD has trouble with tail placement in general, but sometimes it can help to delay the drawing of the tail by using [:tiger tail:0.2] or similar. Adding modifier tokens like "tiger tail under" or "tiger tail below" has some marginal benefit
It's relatively common for the number of turns in her braids to increase from model, especially for tall images
Generation of views from side and back of her clothing are less consistent, but can work (I specifically added training data to get it better at this but it's pretty limited still)
Her vol 15 outfit is not supported