This is a LoRA that generates women squeezing their breasts into one another. It's almost akin to paizuri, but 1girl and solo. It doesn't support breast sizes smaller than "medium" given the nature of the concept, either. I plan to make clothed/unclothed generation consistent and increase the quality of hand generation in the future. Until then, I hope you enjoy!
V1.0
The model's in a pretty good place now. got it back to where a weight of 1.0 seems to work fine, generation consistency is pretty spot on, and it will even generate some side views if you increase the weight of the view tags.
V0.2
Improved the model overall: better hands, better tagging, better breasts shape/squish, etc. Seems to work best at a weight of 0.6 now, but feel free to play around with the weight to get what you're looking for. Weight changes don't throw everything off, so you have freedom with it. If you do still run into problem with too many/little fingers or bad hands, try this flowchart to fix it:
Add extra digits to negative -> Add bad hands to negative -> decrease the LoRA weight
To do:Better hand generation
Support enormous breast size
Support different levels of squish via prompts instead of weightConsistent clothed breast generationBetter nipple generation
Description
Initial release
FAQ
Comments (3)
I know it's hard to crush small breasts. But it can be done if you play with the power of the Lora and adjust and make reading the tag tricky.
Example:
<lora:BSQv01:0.7>squeezing her (breasts:0.1) together
1. instead of using the full tag "breasts squeezed together" you break it down a bit but keep the words (maybe it will also work by lowering the weight of the sentence)
2. If you make the potency of "breasts" 0.1 you will refer to the breasts without enlarging them.
It may not work 100% (due to the weight of the lora and the tags) but it depends on the scene and the prompt you have. Then you simply adjust the weights and that's it.
Note: This has worked for me with a realistic model. I haven't tried it with an anime-style model.
this would be nice, but it whitewashes my characters no matter now hard i try to prevent it :(
I have another version I plan on releasing soon for both SD and SDXL. The first dataset was pretty small and that seems to be a common problem with the smaller models I've tried.