The training images are entirely synthetic, generated with some flavor of Illustrious, initially, and then put through my typical training phase of producing a Chroma LoRA first, in order to use its prompting mechanism to generate a better dataset, and then targeting the intended model, which sometimes turns out to be Chroma as well.The Qwen version operates well between weight 0.4 and 0.6.To make sure it wasn't over cooked I tested at weight 0.8 and prompted for something that was unrelated to what is expected to be produced by the LoRA. It performed as requested.This model is intended to produce images of a woman manipulating a penis, either attached to a subject or protruding from some surface, with the potential to produce ejaculate.Possible poses that appear to be consistent are standing side by side and kneeling, one or both, facing each other, both with some angle usually.

Files with SHA256: 2135f9bec279f5f8b97c2dd2164bd22f66d94b89dd24470847396c68510674da