Tablet menu

Kandinsky 2.2: all about the updated neural network that creates images


Kandinsky is a neural network designed to generate images. It has several versions, the latest of which is Kandinsky 2.2. The latest version differs from the previous ones in improved performance, extended functionality and more advanced algorithms. Thanks to this neural network, you can get sufficiently detailed images for free on request or the original picture.

The updated Kandinsky 2.2 was released in April. Now the neural network can create paintings in a variety of styles. For example, in the style of cyberpunk, anime, medieval painting, renaissance, realistic photo, etc. The neural network toolkit allows you to use text queries, original pictures, text and pictures to generate an image. All images in the article were made by text requests by a neural network. 

Innovations Kandinsky 2.2

One of the main features of the neural network is the ability to create photorealistic images with high quality. The new version also offers a choice of several image formats: 16:9, 9:16, 3:2 and 2:3.

In addition, the neural network can now generate stickers, which makes it an even more versatile tool for creating various images.

The new version also added a ControlNet structural block, which allows you to change only part of the created image, leaving the original composition. 


If you want to try out Kandinsky for creating images, there are several sources where you can find this neural network. Here are some of them:

Telegram bot



Website "Sber"


To generate a picture using a neural network, you will need to write a query or prompt. The Sberbank website provides a formula that is simple - object + background + style. However, for best results, the query should contain as many details and specific items as possible, not just abstract concepts.

Last modified on
back to top