This neural network creates SURREAL images using words! (PICS)

Russian cosmonauts on Mars
The resulting images seem... strange, to say the least. But Russian Internet users are delighted to share these funny “drawings” on social networks.

Russian financial ecosystem Sber has unveiled the first multimodal neural network called ‘ruDALL-E’, which creates images based on text descriptions in Russian and English.

The creation process is as follows: First, one neural network processes a text query and generates pictures, the second selects pictures that are more appropriate to the query, the third increases their size without loss of quality.

To get a picture, you need to go to or use a neural network bot in Telegram, enter the description of the desired image and wait. On average, the image appears in a few minutes. You can make an unlimited number of requests; each time the neural network will return a new result, even if the query stays the same. 

The neural network continues to learn from pictures and texts. Sber emphasizes that the neural network has great potential - it can be used to create interior design, stock images, images for advertising, copywriting, as well as architectural and industrial design, according to its developers. 

However, some of the images are too funny or creepy and Russian-speaking users of social networks are only too happy to share their results. 

This is what the neural network thinks “Kotik Kommunist” (“Communist Cat”) looks like:

“Yekarnyy Babai” (a euphemism for a swear word):

“Kovidnye Ogranichenie” (“Anti-Covid restrictions”):

ABBA music band:

“Shedevr Sovremennogo Iskusstva” (“A masterpiece of modern art”):

“Ministerskvo Lyubvi” (“Ministry of Love”):

“Voskhititelnaya pustota” (“Delightful emptiness”):

“Inoplanetyane” (“Aliens”): 

“Lenin v galstuke” (“Lenin with a tie”):

And this is what the neural network thinks the Russia Beyond logo looks like

If using any of Russia Beyond's content, partly or in full, always provide an active hyperlink to the original material.

Read more

This website uses cookies. Click here to find out more.

Accept cookies