v3.0 - early release, wait for more specific pose versions and further refinement. Tons of known issues.
768*512, horizontal, 3:2
512*768
576*576
More flexible, larger,
FT15ci
fp32
<lora:ossmsnfgFT15:1>, {ossmsnfgft, }{1girl, }{realistic, }{lying, }{hetero, }{1boy, }{on side, }{sex, }{penis, }{pussy, }{pubic hair, |}{erection, |}{testicles, ||}{completely nude, ||}{vaginal, |anal, }{male pubic hair, |||}{spread legs, |||}{open mouth, |||}{female pubic hair, ||||}{navel, ||||}{ass, |||||}{looking at viewer, |||||}{tushy, |||||}{feet, |||||}{dark skin, |||||}{dark-skinned male, }{interracial, |||||}{leg lift, ||||||}{sex from behind, ||||||}{parted lips, ||||||}{ass grab, ||||||}{facial hair, ||||||}{leg up, ||||||}{leg hair, ||||||}{solo focus, ||||||}{deeper, |||||||}{spooning, |||||||}{toes, |||||||}{blacked, |||||||}{very dark skin, |||||||}{chest hair, |||||||}{arm hair, |||||||}{beard, |||||||}{hand on own ass, |||||||}{clitoral hood, |||||||}
Version 2
768*512 sd 1.5
High model diversity/flexibility
Strong consistency of position
Prompt str 0.6-1
VERY POWERFUL sd1.5 model
Analog madness 7 recommended base model.
*Vaginal and anal available inside, see other models holes not covered here.
<lora:ossplnskFT15rs4:0.8>{ossplnsk, }{1girl, }{1boy, }{on side, |}{on back, |}{hetero, }{pussy, }{penis, }{lying, }{realistic, }{spread legs, }{sex, }{nude, }{completely nude, ||}{pubic hair, ||}{testicles, |||}{photorealistic, |||}{jewelry, ||||}{female pubic hair, |||||}{anal, | vaginal,}{open mouth, |||||}{navel, |||||}{male pubic hair, ||||||}{solo focus, |||||||}{bed, |||||||}{on bed, |||||||}{barefoot, ||||||||}{closed eyes, ||||||||}{dark skin, ||||||||}{interracial, |||||||||}{looking at viewer, |||||||||}{feet, |||||||||}{anus, ||||||||||}{dark-skinned male, }
Description
made with analogMadness_v40.safetensors as basemodel/checkpoint and
colab.research.google.com/github/Linaqruf/kohya-trainer/blob/main/kohya-LoRA-dreambooth.ipynb
FAQ
Comments (2)
You fucked up the lora by including pictures of 2 girls in the dataset. Any time you get lucky and there's 1 image resembling the desired output, you get 2 heads on one body.
If it's your first time experimenting with loras make it simple, don't include variations that you don't know how to control through the prompt training
I like the idea though!
Thanks,
I really suspected that would have been the case, I could have gotten a good 100-150 of almost the exact position. In the end I opted to add all the two girl ones and also the ones with the man from the left and from the right. I will remake it once someone tells me what to do with the epochs and tags properly.
am I also supposed to just annotate all of them too....?





