This is the first time I've trained a model using screen shots from a game. I've included the data set and yes I was very lazy. I actually wanted to test out "Fluxgym" so this is also the first time using that.

I wanted to not only capture the face but capture the body. As such a lot of the time she'll generate with some variation of the default skin in game. Due to the training set with her "hand up in the air" a lot of generations will put her hand in the air, but not always.

I was kind of surprised to see some "photorealistic" generations here that look pretty good.

I have not tested any LoRA strength. All the images generated from this are strength 1.0