It was trained on 144 photographs using absolute reality model. It also works with 2d and 3d models. You can use between 0.6 and 0.9 depending on the checkpoint you choose. It is not very successful in text and numbers written with balloons, but I am sure that you can create fun pictures.