Image edit model! The model may sometimes incorrectly generate images for a side view, or it may not generate a back view at all when the chest is hidden behind the body. There are few strapless swimsuits in the dataset - they can be mistaken for other clothes or lack of clothes.
The LoRA is trying to determine if there are proper clothes, and then enlarge the breasts without cleavage (which turned out to be the most difficult part and does not always work)
Description
Works great with 2511 lightning lora.
May work somehow with 2509 and 2509 nunchaku but it will do nudity much more often.
Works with QuantFunc's nunchaku 2511 use "best quality" ver + lora instead of "ultimate speed"
FAQ
Comments (6)
It works really well with 2511, thank you
Is this supposed to work with a mask as well or only on full images? I'm using this with Qwen Edit 2511 but when I edit with a mask, there are no changes at all to the original.
I suppose it's only a full image. This lora is trained on was-became pairs, not on generating from scratch. To be honest, I haven't tried using masks in qwen edit or flux klein, only in the generation models, not the editing ones, because the point is that the model analyzes the image itself. I can post the first version for klein, which is trained for generation, but it almost ignores what is being dressed.
@sabin that makes sense, thanks
效果非常棒。出乎意料的棒！完美的弥补了我作图中的一些需要。感谢创作者！
tag your raceplay kink