This is a LoRA trained on about 100 images of Negev from Girls Frontline, the results are quite mixed and likely I didn't use the best training parameters but it's at least better than base model.

I had to ask Gemini-cli to modify SD-scripts to even get the training to work without OOMing immediately on my crappy AMD GPU, and this took a whole day because AMD is garbage so have some understanding. I might try training a model over a few days (no it really shouldn't take that long but alas everything is NVIDIA focused) with better parameters if I can pause and resume training.

Finally, no I am not a xyz political ideology she is just cute now leave me alone