Trained on 80 hand-picked images with Rank 32 at 512, 768 and 1024 resolution.
This time I used JoyCaption, which needed relatively little correction, though correcting 80 images still took a while :P
The dataset included:
Blowjobs, handjobs, vaginal sex, anal sex
One woman with two dicks, two women + one penis, and two women with two different dicks
Side view, from above, from below, and from behind
Close-ups and wider shots
Various rooms and lighting conditions
It transfers well to many different ideas and art styles.
Enjoy! If you can remember, it would be awesome to see your creations when you tag the LoRA.
I used the new validation loss graph in AI-Toolkit to find the point where the LoRA starts to overfit. It works pretty well. I trained up to 6500 steps, but after a lot of testing the sweet spot was around 5000 steps, even though the graph kept decreasing a bit after that. The validation loss is calculated on images that were not part of the training data (I used three different ones).
Description
FAQ
Comments (5)
fellas, this is a great lora cuz it proves that it sure is a girl on the other end of that glory hole! yup, that's 100% female mouth and hole, and you can tell by them girly noises she makes with the occasional deep moan! and don't let anyone tell ya different or let that quick glance of a goatee fool ya!💁♀️
He's right, I feel vindicated
a hole is a hole lol
i would be more worried about stds and stuff than whose warm wet mouth i'm in
@Protonium just make sure when you run the model on your computer that you have a good up to date antivirus PrEP installed and your models will be fine loll
I generate, because I know the real truth of the Gloryhole...
If I stick my cock there, someone is gunna snap a mousetrap on it.