As someone who does real art and who has also fiddled around with AI. I'd like to say something.
The effort it took to train this model is nothing compared to the effort it takes to learn, study, and master art. Anything created with this model should be created only for novelty and not for profit or validation. As it takes about 10-60 seconds to generate an image with this, Anything created with this model is inherently worthless. No amount of epochs, samples, batch sizes etc will even replace genuine art.
You aren't special or talented for using AI. The results might look good, but it's just for fun. Please support real artists.

Generates headscissor thigh choking poses where the opponents head is facing towards the aggressor. Sort of like a cunnilingus pose with extra squeezing.

Trained on 226 images, with a very slow learning rate over 5 hours. Hopefully that results in it being a bit more flexible.

All the example images are there to show what you can get with just prompting (no control net or in painting required.)

Prompt guide:

There's a few tags I used in the data set to differentiate the poses of the two characters.

on side, on back, lying on person, sitting on person, standing, lying on ground

Some things you might want to add to the negative prompts:

watermelon, mask, wrestling mask

I found on side works the best, but it's quite flexible with different poses and camera angles.

If you found this model useful, please support me on fanbox: https://arturitwo.fanbox.cc/