Reprint, not my work. Original description:
Trained with smaller dataset of previous lora for 1.5 model, based on Pony XL.
v03, cleaned up the dataset, more focused training.
Just in case of confusion, you won't need to put all triggers in to get the result, main thing you'll need is "spread *****" and "cervix". the rest is to enhance the experience, and "close-up" is for, well, close-up shot.
Cervix and urethra have better shape and has less style influence for the style (still has some), work with style loras better comparing to the previous version.
Dataset contains 99% of anime style imgs, if this can do realistic, it's a blessing, if can't, you know the reason.
I still don't have a clue how to prompt on PonyXL yet, so the img quality is pretty unstable... please excuse me.