Flat chastity cage + cum. Have a nice day.
DO NOT TIP BUZZ, I DON'T USE IT. THANKS.
Description
Trained on 300 flat chastity cage photos of various colors. No faces in dataset, ensuring likeness.
Front works well. Not so much from behind.
Prodigy linear balanced 5500 steps.
FAQ
Comments (2)
I don't know what you did training-wise with this one but it works phenomenally well.
My only criticism is that it always adds cum.
I also noticed that it produces probably the most accurately rendered testicles of any LoRA I've used. And if you put it at a decent strength but just DON'T prompt for the chastity cage you can still get them rendered nicely, although it does then add cum.
If you're looking at doing any further in future with a similar training approach, I'd humbly request no cum (for flexibility) and PLEASE consider a small penis LoRA. Everything we've got right not for trans/femboy content is massive swinging dicks and completely neglects the feminine/flaccid smol pp.
Anyway, great work!
Data contains clean chastity images without cum and was captioned meticulously. But 95% do contain cum. I'd recommend using a prompt strength node like kjnode (krea prompt weight node) to negative prompt (cum:-1).
I am considering small flaccid uncircumcised pp training, but just can't find a good enough dataset to my standards. Most nice ones are LQ anime pps, and HQ ones are ugly slightly too big/flawed sissy pps. If you have a source or suggestion I can look.