Hands on with ‘Vincent’, the AI that creates ‘works of art’ from simple sketches
Simple, hand-drawn sketches are being converted into paintings reminiscent of some of the world’s most celebrated artists using ‘Vincent’, an artificially intelligent (AI) system based on machine-learning algorithms.
Developed by UK-based firm Cambridge Consultants, Vincent is capable of creating paintings from even the most rudimentary line drawings.
After its initial unveiling last month, the company said that “much of what makes Vincent tick was not known to the machine-learning community just a year ago”.
8,000 pictures from 1,000 different artists, including Rembrandt, Picasso and Van Gogh, were fed into the system.
The developers included everything from renaissance-era old masters up to the modern day, but most of the compositions produced by Vincent appeared to emulate an early twentieth-century impressionistic style.
Although human judgements on art are largely subjective, the artists that the algorithm creates its images from are some of the most popular in the western world.
A simple, child-like line drawing of a landscape consisting of little more than a line for the horizon and a basic tree produced a remarkably aesthetic composition.
The system did not work so well when presented with a smiley face. A Cambridge Consultants representative noted that this kind of simplistic imagery was not typical of the work by the kind of artists from which Vincent’s dataset was drawn.
The machine-learning process uses both a simulated ‘painter’ and an ‘art critic’ to improve the results. The critic tries to determine whether the compositions were original or created by Vincent itself.
This back and forth process resulted in the painter producing increasingly impressive pieces in order to fool the ‘critic’, which in turn became better at discerning fakes from the real thing.
Different styles could be applied to the line drawings sketched by the user, but although they differed in approach, the aforementioned impressionistic art style was consistent throughout the results.
The project was demonstrated at Nvidia’s GTC conference in Munich as it uses the company’s graphics cards to process the machine-learning algorithms.
Cambridge Consultants has also been working on a machine-learning algorithm that can determine different types of piano music (ragtime, jazz, contemporary, classical, minimalist etc).
The firm said that this bot often struggled to remember the overall tone of a song, getting confused about its genre during passages that may differ from earlier movements.
In July, Neuroscientists at the University of Washington developed a hands-free musical instrument, the Encephalophone, entirely controlled using the mind.