LoRA for generating images with the "Breast Press" pose
LoRA request queue on my discord: https://discord.gg/DAfjhRfyEM
V2.0
This model was trained using 248 images of different characters with varying breast sizes in the pose. It was trained on the NAI model so it should work with most anime models.
Recommended weight is 1.
Examples were generated on JavaMix with 30 steps on euler a, hires fix by 1.5 on 0.4 denoising and adetailer for the faces (face_yolov8n)
Note: This was made exclusively on anime images and on an anime checkpoint. Feel free to post images on realistic, semirealistic etc. checkpoints, but do not rate it on that! This was made to be used on anime checkpoints so keep reviews for that!
V1.0
This model was trained using 110 images on the NAI model. Most of them were NSFW, so that's what you will mainly get unless you specify it otherwise. I only tried it with 2 girls so far, might also work for futa or with a guy.
Works best by itself but with some inpainting you can get it to work with other character LoRA's.
I got the best results with a weight between 0.6 and 0.75.
Trigger word is breast press but I recommend adding 2girls to make it consistent when using other LoRA's. Add kissing and closed eyes if you want to achieve that pose. Add nipples to the negative prompt if you want the chest area to look cleaner.
Example images were generated using AOM2 hard on Euler A with 20 steps.
EDIT:
Male on female does not work, for futa on female just generate the image normally, then inpaint the futa dick since it also doesnt generate it well on its own. Also I did not clarify this before but it only got trained for the pose in the examples.
Description
FAQ
Comments (8)
How to put 2 specific character ? for ex I want put hinata and Tsunade together
Theres multiple ways to achieve that. Either use something like the regional prompting extension, or just generate a normal image with the lora, then go to inpaint, inpaint one of the two, throw on the first character lora (lets say tsunade) and generate again until you have a good tsunade, then remove the tsunade lora, put on the hinata lora, inpaint the other character and generate again. If the two models are too different you can then at the end use img2img with only this lora at a low denoising strength to make it look cleaner.
Can this be used for double standing doggy sex with 2girls and 2boys? I hope you can develop such concept/pose since this one is really really good.
Probably not, their current position is kinda baked in and they'd have to kinda bend a lot (?) to achieve this with doggy style. Ive browsed and found like 2 images of what you are referring to but I can't make a model with this small sample size. Sorry.
@DerHammer That's too bad. I'll try to look for more of it, might vary but I'm pretty sure some CGI H sets might contain such concept. Does replicating images available using controlnet advisable to be used for training images? Though I'm pretty sure it won't be as accurate as it will be right? Hehe...
@DerHammer Ok, this might help hehe https://www.pixiv.net/en/tags/%E7%94%B7%E5%A5%B3%E5%A5%B3%E7%94%B7/artworks
I noticed that such LoRA exists probably private use. Hope you'll find this helpful
@debridpro438 Do you mean this one? https://civitai.com/models/62754/concept-symmetrical-docking-orgy
@DerHammer Oh yes yes, that one. Have never used Lycoris before, no wonder I missed that one.