Your personal Dali: SberBank launched a neural network that draws pictures by verbal description

The SberBank team announced the launch of the ruDALL-E neural network, which is capable of creating images based on textual descriptions in Russian. As the press service notes, this is the first such neural network in the world.

Your personal Dali: SberBank launched a neural network that draws pictures by verbal description
Official advertising picture of ruDALL-E

Anyone can test ruDALL-E, but you will have to wait a bit. The service immediately warns about this and reports the approximate time until the image is ready. At the time of this writing, it took the service 9 minutes to generate a picture according to the description “Cute cat is reading iPhone”. The result is like this:

Your personal Dali: SberBank launched a neural network that draws pictures by verbal description

The neural network is simultaneously trained on two types of data – pictures and texts, and allows you to create an unlimited number of new images according to a given description.

The ruDALL-E XL model (1.3 billion parameters) is free to use by downloading it from Github.

The creation of images using ruDALL-E occurs in three stages: first, one neural network receives the text and generates a given number of pictures, then the next one chooses which of them are the most successful and most closely match the description, and the third increases them in size without loss of quality. Thus, you can get an unlimited number of new images that fit the specified characteristics.

You can see examples of generated images here.

.

You may also like