>>8670772
I think >>8670868 has a good idea with lipstick in the negatives, since I've seen the AI conflate lips with lipstick before. Running a batch right now but I can test that out once it's done.

>>8670782
Like >>8670788 said, if possible you want to use more than 9 images when training a LORA. Although quality matters more than quantity, having a higher quantity of images help mitigate any issues in any one image and can help the AI learn the character better.

>>8670505
This is a fun gen! Neat that you got both characters to turn out clean and in different clothes without any bleedover.

>>8670283
I have not tried Xinsir yet, but in a previous thread an anon recommend it: https://huggingface.co/xinsir