This LoRA is intended to cause wardrobe malfunctions. It has two activation keywords: downblouse and nipslip
Downblouse produces a view down the subject's shirt, while nipslip permits nipples to escape their prisons. The two keywords can be used together (but don't need to be), and can even be used in negative prompts for more control over the composition.
You can also eschew the trigger words and just increase the weight of the LoRA to encourage wardrobe malfunctions.
IMPORTANT: Version 3.2 was trained at 640x640 pixels and produces best results at that size or larger; it may produce oversaturated results at smaller resolutions. If your hardware confines you to using smaller resolutions, please use version 2 instead of 3.
Description
Revisited the training data and parameters to try to give it a better understanding of the difference between a downblouse and a nipslip. Hopefully it compares favorably to 3.2!
FAQ
Comments (10)
Position of nipples are sometimes weird (too high), but I really have fun with the model. thanks!
Yea I've found the same thing with these types of LORAs, it seems like the model thinks the nipple NEEDS to be fully visible. When in reality there is just a suggestion of the nipple since it's usually pointed down. I think in time we'll get better models. Right now I find myself having to do Photoshop to fix images.
It actually works better for me without the trigger words. Seems to find more natural ways to create malfunctions :)
really tried to make it work, but it's just too inconsistent and the results are usually pretty scuffed
No worries, hope you find what you're looking for! I suspect it just doesn't play as well with revAnimated as it does with photorealistic models.
@Invisidude tried RealisticVision, it was just too inconsistent. I'd crank out 20 images and maybe 3 fit the concept.
DownBlouse Extended gave me the best results so far, but the bias for making a "deep v" makes it incompatible with certain clothing
@pogo Yeah, that makes sense. Probably depends heavily on the pose/clothing choices. I found with RevAnimated I needed to drop "standing, hands behind back" from the prompt to get a typical downblouse pose, but even then it didn't give me the results I've come to expect from other models. I suppose I'm just a fan of the "deep v" haha
I tried to use it for hentai, it doesn't seem to work properly, anyone had any success doing it or know any other lora like this for hentai malfunctioning clothes?
What model are you using? Depending on the training data and captions, your base model may be sensitive to phrases such as "hanging breasts" "nipple slip" "one breast out" and "areola slip". In particular, the models that have Pony Diffusion in their ancestry are good enough with those prompts that no LoRA is typically needed for the effect.
well made
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.