This model tries to get the classic over the knee spanking pose with consistency... did we really need another OTK LoRa? Welp, maybe not really, but I wanted to experiment and I think it works quite well.

This model is trained on two datasets and has two triggers, mostly as an experiment:

over the knee is a strict selection of the classic pose with little variety - sitting, free dangling feet, straight perspective, no extra people or anything

otk_spanking is a looser selection of more permissive posing and perspectives - kneeling, feet and hands on a sofa/floor, etc

The idea here was that essentially combining two datasets could maybe yield better results when used in combination. And indeed, one trigger produces around 25% of what I'd consider failed outputs, and combined it's reduced to around 12,5%... ymmv, though.

You should probably use over the knee, otk_spanking, spanked, spanking in the prompt for best results.

The training data does not include males being spanked, btw, and it seems it's biased towards F/f over M/f.