It may not be immediately clear why someone wants a LoRA for something that the model can produce with some acceptable cohesion, after excessive prompting, but I can explain it very simply. Token efficiency. I'm not sure why someone would spend 150 tokens on an item when 2 are enough with a LoRA so yea, while someone might be the worlds most awesome "Prompt Engineer" I figure taking a shortcut doesn't fall into some sort of moralistic category.The training run was done with 6 images, of those only 1 was real.Some people like the "real" look while I tend to enjoy things that I don't see often, these days, like 3D, airbrush and deep contrast pencil work.This LoRA might accidentally lean towards real, or something that irritates some people enough to feel a need to critique the lighting, in order to feel good about themselves.This can do at least two very different configurations which are a split leg pose and legs behind the head, the latter being more prominently trained and has more fun doohickeys without having to prompt them in.

Files with SHA256: e2f6a2b535a31c2db67f058a18d88d7b08fd1b06863fa571547970f57e6dc197