The hierarchical architecture of GANs can be useful in computer graphics applications and can help reduce the number of labeled examples needed for unsupervised learning. Deep layers and local connectivity of convolutional deep learning can avoid the "curse" in this kind of modeling.