At the GTC 2019 conference, taking place these days in San Francisco, Nvidia presented the GauGAN neural network, trained to turn primitive sketches into photorealistic images. It is reported by the TechCrunch resource.
The new development from Nvidia formed the basis of a graphics editor containing only three tools – brush, pencil and fill. With their help, users can draw simple shapes, assigning them the properties of real objects – clouds, plants, bodies of water, etc. Based on this data, the system automatically generates landscapes with a sufficiently high degree of detail.
So far, the GauGAN neural network can only create images of landscapes, but Nvidia assures that it will not be difficult to teach it to other genres of fine art.