Trained on a 30-image SDXL 1024x11024 dataset with detailed portraits, full-body poses and high-variety of lightening. Best results at LoRA strength 0.6-0.9