LoRA is generated entirely from anime images from S1 (not a fan of S2 chara design).
-----------------------------------
Works well with Pony models.
Check samples for prompts.
Note:
Using hi.res fix is recommended.
In case you get inconsitent colors, add to the prompt: 'green dress, white dress, orange dress'
EDIT 09 Aug.: I just noticed that for some reason she likes to generate with parted bangs. Place it in negatives to mitigate this issue.
There are literally zero images like that in the dataset and it happens even with the other epochs, sadly... so can't say what went wrong.
-----------------------------------
Leave a like, comment, review but first of all, ENJOY ;)