Ganbreeder uses a machine learning technique called Generative Adversarial Networks (GANs) to generate images that seem like photos, at least a first glance.
GANs use a pair of networks, one of which produces data and the other classifies it; the "generator"'s goal is to get the "discriminator" to identify its data as belonging to a certain class (like a stop-sign, or a fingerprint), while the "discriminator"'s job is to successfully discriminate between real data and the synthetic data created by the generator.
Ganbreeder uses a GAN to create creatures that appear to be objects of various types; you choose your favorite and the GAN produces a bunch of "children" for you to choose from, chasing down a rabbit-hole of weird, machine-learning found imagery. Fun!
(via Four Short Links)