Please read the release notes over there 👉 or up there 👆 if you're on mobile.
Version 1.0 gallery images were made with the base SDXL 1.0 model.
Version 2.0 gallery images were made with RealVis 3.0 and the base SDXL 1.0 model, then touched up with some inpainting.
Version 3.0 will have more comprehensive tagging to control pubic hair.
Try these resolutions:
1024x1024
896x1152
768x1344
1344x768
1152x896
1024x1024
...and others!
Description
Please see gallery images for prompt ideas.
Hard to control caucasian versus black cock. Also hard to control facial expression. These will be fixed in the next version.
If you put "open mouth" in the negative, make sure you increase the weight of "reverse cowgirl" in the positive prompt (eg: (reverse cowgirl:1.5)) - but play with the weights. A lot of the training data had women with open mouths so you may find this necessary if lots of your generations have an open mouth expression.
Trained with the following captions
photo of a woman, reverse cowgirl
photo of a woman, wide shot, reverse cowgirl
photo of a woman, reverse cowgirl, closeup
Supported resolutions
832 x 1248
768 x 1280
1248 x 832
1280 x 760
1024 x 1024
...please try others!
FAQ
Comments (4)
Great work.
Does it work well with your NFSW model? Any general BG model being worked on or that you recommend that keeps things realistic and classy instead of plastic and thrashy?
It seems to work decently well with Realvis XL and Juggernaut XL! I do have a reverse cowgirl 2.0 I've been meaning to post.
@MachineMindedÂ
Sounds great. Appreciate all the hard work you do good man.
Do you have a good estimate how much work it would be to train a single lora or checkpoint to include everything from couples strolling in the part to stripping down for hardcore sex on the hotel room bed?
@ubit34Â It could probably be done but is a little out of my wheelhouse. I've tried blending concepts in LoRAs before and it doesn't work that well, unless there is a fundamental thing I'm missing.