Until recently, the generation of images in computers was mainly done via algorithms that leverage mathematical rules of physical laws, i.e., are
relatively simple to implement. What remained problematic for a long period of time was the generation of images that closely reassemble hand-drawn artwork.
The critical issue is that unique artistic style is typically hard to describe
by a set of mathematical rules. Nowadays, in the age of artificial intelligence where cars are nearly autonomous, and computers can defeat us in chess or Go, one could expect AI to generate art indistinguishable from canvases of famous painters. Surprisingly, the first approach that passed the artistic version of the Turing Test was not based on neural networks. In this talk, we discuss why that happened, the current capabilities of neural techniques, and how they can help artists produce unique artistic content automatically in near future.