Trained on a very very small dataset of 12 images with regularization. Trained until full body image gave consistent resemblance, contains residual artist's style.