The GauGAN 2.0 neural network from Nvidia has learned to draw pictures by verbal description

Nvidia developed and showed artificial intelligence to create photorealistic images called GauGAN back in 2019. But the neural network has only recently acquired the form of a full-fledged tool available to the general public in the form of a Canvas application. And now the GauGAN 2.0 version has been released, which now knows how to recognize verbal descriptions of what you want to draw.

The main feature of GauGAN is not only in recognizing the essence of user requests, but also attention to the details that they would like to reflect. Ideally, it can change the shape, size and texture of any object in the drawing arbitrarily based on text and graphic instructions. And at the same time preserve the overall harmony and integrity of the canvas, which eventually looks like a photograph or an artful painting.

In order for the AI to understand human needs so subtly, the generative-competitive model was trained on the examples of 10 million different landscapes. Therefore, she understands the difference between a “mud-dusted boulder on the shore” and “rolling stones in the surf”, and can draw both in one frame. What is even more interesting, the changes are displayed in real time, as the query is compiled.