The RePrint system, under AI control, learns to forge master paintings at the highest level.
Forging paintings by famous artists is a long, one might say old craft that brings a lot of money. Some fakes are so good that art historians have been arguing for years about whether it is a fake or the real canvas of a master. Now a team of scientists from MIT is developing a computer system that in the future will be able to forge paintings at the highest level.
The idea is that the computer “painted pictures” so that not only the details were indistinguishable, but even the shape and depth of the brush strokes remained. So far, the project, called RePrint, is far from ideal. He paints with a special paint that is suitable for a 3D printer. Neither about oil, nor about a water color or other types of paint while and the speech cannot be. But who knows to what heights of skill AI will reach in the next few years?
So, the system in question consists of two parts - a 3D printer that applies layers of specialized ink, and a neural network that controls this printer. In order to recreate a copy of the picture, neural networks "show" the real canvas. The system evaluates the appearance of the picture, adjusts the light palette, evaluates the color gamut, which should be present in the picture.
Next comes the printer. The main difficulty is that the ink is applied in such a way that it creates the impression of a real web, including strokes and even defects. Small problems are almost always present, and often they are the ones that indicate that the canvas is the original, and not a fake. The developers used machine learning, so the system gradually learns from real artists.
Interestingly, the computer shows photos of paintings, and not real canvases. And the output of the system creates a fairly accurate copy of the original. Of course, the difference in the lighting of the picture during its photographing, the clarity of the photograph and many other important points play their part.
Source: mit.edu
Above shows the original (left) and its copy, created by the machine. The difference is quite noticeable, unfortunately, the master’s brush strokes have not yet been learned to reproduce the machine. But over time, the developers hope to achieve that. Now at least the color scheme is very similar, there are especially strong blunders (with certain reservations not).
The advantages of a system that “forges” canvases are obvious — after all, it can reproduce both the three-dimensional structure of a picture and give a more extensive color gamut. The 3D printer works with 10 colors and shades instead of the usual four - blue, red, yellow and black. The system is taught to “understand” the colors thanks to the old technique, which consists in applying a lot of colored dots instead of creating a picture using brush strokes. Combining technology and old technology allows, if not to perform miracles, then get close to the original technology.
The neural network that controls the entire process of recreating a picture was trained on the example of 18,000 real canvases. With regard to the physical structure of the canvas, the developers in the near future hope to increase the resolution of the print head, as well as better work out the color gamut that the neural network reproduces. The fact is that some colors are not yet available for a 3D printer. This applies, for example, cobalt blue.
As for the purpose for which the “virtual artist” is being created, this is by no means fake paintings. On the contrary, the authors of the project hope to show lovers of art paintings who, for whatever reasons, gather dust in the storerooms of museums, have been stolen or inaccessible for other reasons.
It is worth noting that the idea of scientists is not new. Back in 2015 it was reportedabout the start-up Relief Art, which presented a special way of printing Dar3D, which allows you to convey the color, pattern of the picture, as well as the amount of paint on it. According to the developers, this technology allows you to put an image on paper, canvas, leather, wood and even polyethylene. The technique of creating paints volume on the canvas is called impasto, it allows you to give the image both depth and expressiveness.