The new Kandinsky 2.1 neural network from Sberbank: creating unique drawings in Russian.

Instructions for using Kandinsky 2.1

Kandinsky 2.1 is a neural network capable of creating images from scratch and processing existing ones. It was developed by the Sber AI team, a division of Sberbank that creates services based on artificial intelligence. The neural network is available for free on Sberbank's website and on several other platforms.

Kandinsky 2.1 can create images based on text descriptions. For example, if you write "a person is reading a book", the neural network will create a corresponding image. In addition, it can be used to edit existing images, such as merging multiple images, removing backgrounds, or adding missing elements. The neural network also allows you to choose the style of the images you create or modify, for example, the style of Malevich or Soviet cartoons.

Kandinsky 2.1 is an improved version of Kandinsky 2.0, which was trained on one billion text-image pairs. Kandinsky 2.1 has been trained on 172 million high-resolution images, including high-quality images with text and human faces.
Neural network can be used in different industries such as marketing, SMM, design and advertising. For example, it can be used to create illustrations for posts in social networks, as well as to create personalized marketing solutions. The neural network is available without restrictions and does not require registration on Sberbank's website. It is also available on Sber smart devices, in the Salyut mobile app, on ML Space, Fusion Brain platforms and in the Telegram bot.

How to work with the Kandinsky 2.1 neural network on the Fusion Brain platform in a browser: step-by-step instructions

No special knowledge is required to work with Kandinsky 2.1 on the Fusion Brain platform - the interface of the device is quite simple. To create an image, you need to describe its content and click the "Create" button. You can also upload your own image and write instructions for processing it.

You can set the image style in the description or choose from the offered options by clicking the "No style" button in the lower left corner. If you change image styles and description, the neural network will create new results each time.

The image can be modified - crop the desired area, erase part of the picture with an eraser - or downloaded by clicking on the download button in the upper right corner. The picture will be saved in PNG format.

It takes about 30 seconds to process a request to create one image. In case of high load on the neural network, the processing time may increase.