Microsoft has developed Gen Studio, which can create virtual art using artificial intelligence in cooperation with the Metropolitan Museum of Art in New York and MIT. AI is released through the website so that anyone can freely enjoy art in their imagination.
The Metropolitan Museum of Art has released more than 375,000 art works images held in 2017 for free, allowing anyone to freely view or use the works of the museum. This year, the Metropolitan Museum of Art celebrated the second year of the project and held a hackathon for two days in collaboration with Microsoft and MIT. It aims to develop AI that can create virtual art images by combining the image data released by the Metropolitan Museum of Art through the hackathon with AI technology.
The artificial intelligence algorithm used in the project is a kind of generative adversarial network. GAN uses two neural network systems. When a neural network creates a virtual art image through a generator, a discriminator determines whether the art image created by the competing neural network is correct.
Through this, if the work of art is judged to be fake, the generator evolves to create a more accurate art image, and on the contrary, the discriminator gradually increases the judgment accuracy of the image. By repeating this process, the precision of artificial intelligence for creating virtual art increases.
The Zen studio created in this way is open to the public, so anyone can enjoy art created by artificial intelligence for free. When you access the site, you see a large number of images across the screen. All of these images are created by artificial intelligence. You can also select an art type by clicking the category selection menu in the upper right. Whether it’s an accuracy issue, there is no category like sculpture or painting. The generated image can also be saved as a separate file. Also, if you click Search Similar Images, you can view images of real collections close to those created by artificial intelligence.
Zen Studio’s source code is open source and published on GitHub . Related information can be found here .