Pixie

Trained on a 26-image SDXL 1024 x 1024 dataset with detailed portraits, full-body poses and high-variety lighting. Best results at LoRA strength 0.6-0.9