Trained with at 1024x1024 with hand cropped patches from 20MP images
High detail with broad concept reinforcement