The model is a generative adversarial network (GAN) trained on a dataset of photographs of the French singer Daniel Balavoine. The generator portion of the model creates new 512x512 pixel images that resemble Balavoine's face and hair.

the generator learns to produce varied but realistic looking portraits of Balavoine in different poses, expressions and lighting conditions. The resulting model can be used to automatically generate unlimited novel portraits of the singer with random sampling of the noise vector.