Trained on images of mostly women (though some men) in revealing circumstances and avoiding eye contact. It works sort of well, but may take a couple of gens to get what you want, looking to the side can help in some prompts as well. Works better when phrases like from above, from side, or looking at viewer are avoided and when generating characters in more nsfw situations. Works great when inpainting, especially since SD likes to center eyes by default.
Description
FAQ
Comments (5)
god thank you, it can be so hard to get characters to not stare directly at the viewer!
anime only?
It was trained on anime images and an anime model, so it leans towards that direction. I did some tests using the Latin Beauty model and got some levels of success, but it definitely won't perform as well as with an anime model.
great work, thanks a lot, hope you make lora for eye contact so if there are two people in the photo they look to each other, this will be great 🤞🤞🤞🤞
Can adding negative weight or putting some words in negative prompt help to achieve the opposite effect (looking at viewer)?