Version 1.1 update: Attempted a new strategy for training to try to increase the quality as much as possible (without renting something to train on HD video). I think it worked decently, let me know what you think. Check the notes below for details on the training strategy if you're interested.
The preview videos were generated with a strength of 1. The motion works if you lower it, but it might start to know less about what a penis is supposed to look like.
I included the workflow used to generate the preview videos in the "Training Images" download. I separately ran the videos through film VFI afterward to increase the frame rate.
This was trained on 4 different angles, here is how to trigger them:
A woman is lying on her stomach between the legs of the viewer and performing oral sex on a man. Her head moves up and down as she sucks the penis.
An overhead view of a woman kneeling between the legs of the viewer and performing oral sex on a man. She moves her head back and forth as she sucks the penis.
A woman is leaning over a man positioned in between the legs of the viewer and performing oral sex on a man. Her head moves up and down as she sucks the penis.
A woman is kneeling in between the legs of the viewer and performing oral sex on a man. Her head moves up and down as she sucks the penis.
Training details:
I included a lot more HD images this time, 1:1 with the number of videos. I used 25 each. I trained about 1200 steps, and then checked to see if it had learned the motion. It had, but still didn't have a good idea of what a penis was supposed to look like. At this point I continued the training using HD images ONLY. Since it had already learned the motion, I just needed it to learn what things look like in high detail better. I probably didn't even need to train it with videos for 1200 steps, I likely could have stopped even earlier.
Anyway, the goal was to do the minimum with videos to teach it the motion, but not overtrain on the low res sources. I think the results ended up a decent amount better doing it this way.