A Generative Adversarial Network, or GAN for short, is a fairly recent development in machine learning. Some GANS create the surreal impossible images I imagined the random pixel generator would be capable of. Unlike the pixel generators discussed here, GANS are not capable of an infinite number of images– however their strength is in skipping the static and creating new images that seem to make sense. Their set number of combinations is partially based on the number of images that the machine learning program was trained on.
Below are images created using Ganbreeder. 
