Depending on the model, you can provide various modifiers to get different results. The modifiers work best with the base SD 1.5 model. Photorealistic models will require you to adjust the weights to around 0.5. Other models may "override" how some words behave.
Version 1.0 was trained on 512x768 (portrait) images, so those will result in the best. Version 1.1 was trained on 768x768 images.
You can generate 768x512 (landscape), but you may encounter missing limbs, extra arms, and so on.
Example prompts
legspread a photo of a ????? woman spreading her legs and showing off her ????? 👈 this is always a good starting point
legspread a photo of a ????? woman spreading her legs and showing off her ????? (with labia)
legspread a photo of a (caucasian|black|whatever else) ????? woman spreading her legs and showing off her pussy
legspread a photo of a ????? woman spreading her legs and showing off her (shaved|trimmed|bushy) pussy
legspread a photo of a ????? woman with (small|medium|large) breasts spreading her legs and showing off her pussy
legspread a photo of a (petite|curvy|thick) ????? woman spreading her legs and showing off her pussy
legspread a photo of a ????? woman with (fair|tan|olive|brown|chocolate) skin spreading her legs and showing off her pussy
legspread a photo of a ????? woman spreading her legs and showing off her ????? (and asshole)
legspread a photo of a ????? woman (propping herself up) spreading her legs and showing off her ????? 👈 I didn't have a lot of images for this one, but in the next version there will be
legspread a photo of a ????? woman (with tan lines) spreading her legs and showing off her pussy
You can mix and match these modifiers how you see fit. You can put the other options in the negative to get a better result. No guarantees they'll always work!
Description
Used bmaltais's implementation of kohya_ss LoRA training scripts.
This version is resumed from 1.0 with 768x768 images. It will handle usual 512x768 (portrait) generations well, similar to 1.0, 768x512 (landscape) images may result in missing limbs or other weirdness. 768x768 generations are decent.
It is a little overtrained, so for the base SD 1.5 model, I recommend setting the strength around 0.8, and other models, between 0.3 - 0.7.
Demo images for 1.1 were generated with Realistic Vision 2.0!
Total Images: 116 total images, about half high resolution
Regularization Images: 300 total "sexy woman" regularization images from ProGamerGov's dataset at huggingface.
Trigger word: legspread a photo of a woman spreading her legs and showing off her pussy
Training Steps: ~10000 in 1.1, ~17000 total including 1.0
Unet Training Rate: 5.0e-05 (0.00005)
Text Encoder Training Rate: 5.0e-05 (0.00005)