If you’ve ever dreamed of seeing yourself painted by the likes of Vincent van Gogh, Leonardo da Vinci or Henri Matisse, you might have imagined you’ve missed your chance, but with the development of DeepArt, an algorithm inspired by the human brain, that generates digital paintings by combining your pictures with the artistic styles of other images, your dream could finally become a reality. The algorithm, which was developed at the University of Tübingen in Germany, uses the latest advances in deep learning to process the image data. It first analyses the image, before extracting the key features, such as a face or an object, before digitally painting the new image by repeatedly comparing the initial features of the submitted picture with the painting it seeks to emulate. After around ten minutes, the program delivers a unique work of art.

Perhaps what is most surprising about the technology is the variety of recognisable art styles that are so easily imitated. From the blue and yellow swirls of van Gogh’s Starry Night to the bold lines and shapes of Kandinsky’s Transverse Line, DeepArt’s algorithm is somehow able to interpret the stylistic elements of an image and transpose them onto another with a minimal level of noise. This isn’t any old Photoshop oil painting filter. To the untrained eye, these images, essentially generated by a single algorithm, are indistinguishable from original artworks. Indeed, DeepArt offers a Turing test – a test of a machine’s ability to exhibit intelligent behaviour indistinguishable from that of a human – whereby you can guess which one of two paintings is by a human artist and which one has been generated by DeepArt’s algorithm. On average, people guess just six out of ten pairs of images correctly.


The value of original art

Of course, original art isn’t going to disappear anywhere any time soon. Works of art have long been imitable – most obviously in the form of prints. But this new technology – and whatever follows it – will enable anyone to produce thousands of iterations of personalised works of art using styles that artists may have spent decades refining. And I can’t help but feel that giving people the ability to generate hundreds of unique paintings of their dead cats will somehow debase the styles used. Just as Instagram filters have made photographers of us all by enabling us to convert any dreary old photo into the kind of thing you might find in Amsterdam’s Foam photography museum, so technology threatens to make artists of us all.

But the art industry itself won’t be hurt by the development of technology that emulates human creativity because art economics is fundamentally based on the scarcity of unique objects. What might have once been considered remarkable will be viewed as relatively plain, leading artists to shift their attention to art styles that are less imitable by technology. Proust once said something to the effect of, “we only see beauty when we‘re looking through an ornate gold frame“. The democratising effects of art-generating technology will simply effect a shift of the ornate gold frame. But the development of art-generating technology will also be a boon for artists who want to run more experiments. Just as the development of music production software has enabled composers to experiment with different sounds, so artists will use technology to experiment with materials and textures they might not otherwise have access to.

Create your own

Head over to DeepArt.io to create your own masterpieces. Unfortunately there’s a little waiting time – around nine hours at the time of writing. Having experimented with a few styles, I feel qualified enough to make a few suggestions. First, selfies seem to be more impressive than group shots because of the relative lack of detail in the algorithm’s creations. Second, artworks using a uniform style seem to work best – the recognisable motifs used in paintings like Kandinsky’s Transverse Line, van Gogh’s Starry Night and Hokusai‘s Great Wave off Kanagawa seem to translate particularly well. For more inspiration, click here.