Nvidia introduced a new eDiff-I neural network, which still generates images, but with one feature.

A feature of eDiff-I is that the user writes down what picture he wants to synthesize, matches colors with words, and draws a schematic sketch of the desired result. After a couple of seconds, the neural network will generate an image, taking into account all the nuances specified by the text query. So far, the neural network is not available to all users.