Update 10-15-25: V2 now has 4 targeted poses/positions; each pose was trained on 20 images or more! Again, keep in mind that, even with the improved dataset, this is still very much an experimental LoRA. Additionally, approximately 90% of the dataset consisted of one girl with three boys, so more than three boys will struggle (though it should still be doable; play around with weights, as always). Still, in my testing, it helps stabilize triple penetrations/airtight gangbangs a lot more than native prompting does!

I'll see if I can revisit this to improve upon it (this was made fairly quickly; my initial dataset is only 38 images, mostly images from behind the girl), but the idea is that V1 should help with gangbang/group sex images where oral, anal, and vaginal sex are simultaneously being performed.

Again, V1 will work best with making images from behind and directly from the side. I'd like V2 to have more images of the girl on her back, with her breasts pressed down, etc. But, for now, hopefully this V1 will help stabilize airtight sex images for you!

➼Obligatory self-promotion: consider supporting or commissioning me at https://www.fiverr.com/metalchromex/make-you-a-stable-diffusion-anime-character-model or https://www.patreon.com/MetalChromeX so I can continue making cool crap for you all to enjoy for free!