Wanted to get better results that's not cursed wen only using: 1girl, ass, cameltoe.
so i went on a journey to fix my kohya_ss witch broke on me for some reason.
USE AT 1.0 ,not tested with prompt
<lora:ass-cameltoeV1.0>
V1: trained with 4 images 12 repeats NAi net dim 200 alpha 100 just because i wanted to see what happens. bigger buts and heart shaped
V1.1: trained with 53 images 12 repeats NAi net dim 200 alpha 100. wider size ranges
V1.2: trained with 53 images 12 repeats and text files to try to reduce cat ears errors, NAi net dim 200 alpha 100. wider size ranges
Description
trained on NAI ,dim 200 alpha 100
FAQ
Comments (3)
V1.1 images are using a semi hard to get to Korean model and DDSD
Another term for this is "trefoil" , you might want to add that to the title in case anyone searches for it.
You can find a ton more images for your dataset here: https://gelbooru.com/index.php?page=post&s=list&tags=trefoil
I'd suggest trying to do it with 100+ images maybe (and potentially crop out some of them to focus on just the ass for variety and for detail )
Dear author
This is a LORA for "cameltoe", so why don't you show "pussy"?
Also, is the "<lora:ass-cameltoeV1.0:1>" propt unnecessary? It doesn't seem to be in the sample.
V1.1 has cat ears or intakes that grow out of place. I guess the unintended behavior needs to be fixed.