LoRa trained on 40 images of Scarlett Johansson's face region. Works best for direct generations with resolution under 896x896 and weight between 0.6 and 0.8.