A of researchers at Nvidia released a state of the art generative adversarial network, StyleGAN, at https://github.com/
Each time you refresh the site, the network will generate a new facial image from scratch from a 512 dimensional vector.
They also included pretrained models for cars, cats and bedrooms that you can immediately use.
The new architecture leads to an automatically learned, unsupervised separation of high-level attributes (e.g., pose and identity when trained on human faces) and stochastic variation in the generated images (e.g., freckles, hair), and it enables intuitive, scale-specific control of the synthesis.
You can learn more about the network architecture at a summary a friend kindly spent the time to write up.
More at www.lyrn.ai
source Cornell University