6/29/2026: Hoping to start training a more curated model this week! Stay tuned!
This is my first pass at training a LoRA for Krea2. It already does pretty well out of the box with the rebalance node, but this should help push it the right direction. Tagged with joycaption.
Trained on about 5,800 images from the following "porn" categories:
asian
creampie
doggystyle
ebony
hairy
milf
reverse cowgirl
shaved
spreading
suicide girls
Consider this an alpha/beta of sorts. I'm going to further refine my captioning as well as get more data for a second run in the coming days/week.
Will post more samples in the coming days to hopefully figure out how to use the model better.
Description
FAQ
Comments (5)
Thanks! It seems you're training using LoKR and not LoRA. The file size..
The dimension is 128 - it's not a lokr.
What is the approximate ratio of female to male, in the training images (or the percentages)?
Solo male? Probably near zero.
@MachineMinded It's best to specify all-girl in the title or description. NSFW technically includes 50% of the population (males). This is something Civit should improve. It happens all the time here. They should take instruction from the erotic video industry which would specifically label videos as "all girl" for those looking for that.