An artist spent a decade meticulously studying Vincent Van Gogh, creating a series of imitations in Van Gogh’s distinctive style. In the above painting of the Golden Gate Bridge, we can observe the thick brush strokes, gentle dotting, and vibrant colors that emulate the Starry Night almost perfectly. The artist’s dedication to imitation clearly shows that hard work and attention to detail can truly bring about perfection.
It’s a heartwarming story, but it’s also a complete lie. This picture was created not by an artist, but in two minutes by a computer using a novel style-mixing algorithm. Created and published in 2015, the algorithm allows any desktop to generate these pictures on the fly.
Anyone can now make a masterpiece by specifying two images to the algorithm: one a picture and the other an artist’s work. The algorithm then proceeds to extract the style from the art piece and apply it to the picture, generating an image that appears to be meticulously created by the original artist.
The algorithm to generate these images only became possible because of recent computer vision breakthroughs inspired by the human brain. The algorithm that creates these images is based on a piece of software called a “convolutional neural network”, which is modeled to replicate the way in which our brains process visual information. These neural nets are currently being used towards a variety of applications, from state of the art image recognition to reinforcement learning.
Currently the image generation for style mixing is not perfect and images often end up with odd artifacts and artistic deformities, but the process is only improving.
The neural networks that create these style mixes are “trained” using millions of images, but as the hardware behind them improves, they will use billions. This magnitude of difference will allow the nets to create stylized pictures more precise, more intricate, and more beautiful than the ones that can currently be made.
This marks the era of the next evolution of art; a step that leverages sophisticated machine learning as a tool to help us create in a new and different space we never could have imagined. Style mixing is but the tip of the iceberg, for in the coming decades, every modality of art will be affected by machine learning, from drawing to photography to writing. Drawings soon will be become walking 3D models, your short stories fully animated films, your photos immersive 3d scenes. The future will be defined even more by human creativity set free by machine intelligence, so stick around reader, the coming years will be more beautiful than you can possibly imagine.