We use a variation of Generative Adversarial Networks. Our Generator learns separately about style and content, which allows it to interpolate between styles and mix style and content in novel ways.
A "Critic" trying to get better at detecting "fake" images, and a generator trying to get better in fooling the Critic. The Generator improves and advances through this process and ultimately learns how to generate images that pass as "real" art in the eyes of the Critic.
Over time as our network becomes more intelligent we will be able to generate pieces with an exponentially improving 'brain', creating even more creative and beautiful pieces.