>>105777589
I've also been testing it on some anime images as well, so the lora just might not be translating too well onto those. It also recognizes "topless" in the prompt like 20% of the time, but I'm pretty sure you haven't captioned for that.

Given that it has changed the boob size of some of the images I tested, it makes me think it would be nice to see some sort of breast slider lora specifically for Kontext, but that would be a whole different undertaking.