Daniel Sýkora

Can AI paint like an artist?

Artistic style transfer is nowadays surfing on a huge wave of AI hype. The overall impression is that ubiquitous generative adversarial networks can quickly solve almost any appearance translation task, and thus machines could be easily able to draw in the style of famous artists. However, what if we look carefully behind the curtain and put off the pink glasses of hype-generated excitement. Can the machine really pass an artistic version of a Turing test? In this talk, we look closely into this question and demonstrate that (“Hold on to your papers!”) we are still quite far from that point in general case. We will discuss algorithms able to deliver results that can almost pass in some particular circumstances; however, those in their core are surprisingly not based on neural networks. Why is that? Would this change in near future?


Daniel Sýkora is a Professor at the Department of Computer Graphics and Interaction, Faculty of Electrical Engineering, Czech Technical University in Prague where he leads a research group focused on developed of algorithms for artists. Their goal is to eliminate repetitive and time-consuming tasks while being able preserve uniqueness of handcrafted artwork and provide full creative freedom. Daniel and his research team collaborate with renowned industrial partners including Google, Snap, Adobe, Disney, or TVPaint Development to integrate algorithms they develop into professional tools and put them in the hands of artists. For his work, Daniel received numerous scientific awards including prestigious Günter Enderle Best Paper Award and The Neuron Award for Promising Young Scientists.

MartinDaniel Sýkora