In one day since the release of the new version of the Sber Kandinsky 2.2 neural network, users have created 1 million unique images, the press service of the bank reported.
It was noted that the most popular subjects were “cats”, “sea” and “anime”.
“With Kandinsky 2.2, users are able to create high resolution photorealistic images, the difference from the previous version is especially noticeable in portraits. And if previously the pictures were only square, now it is possible to adjust the aspect ratio during rendering. The neural network shows how to create unique stickers in Telegram, where you can combine sticker packs. He learned too.”
In addition, the Kandinsky 2.2 bot in Telegram can quickly replace individual objects or elements in the uploaded image according to the text description, thanks to its ControlNet architecture. At the same time, it will preserve the composition of the original picture.
Also, the new version of the Sber neural network can create an illustration from several images, shape a drawing according to a text description, create pictures similar to the ones given, complete missing parts (internal painting), draw in infinite canvas mode (painting).
“The Sber neural network, unlike its foreign counterparts, understands requests in Russian. It can create images in more than 20 styles,” the message says.
As clarified in Sberbank, it can be tested on Kandinsky 2.2. promotion page models are available on the FusionBrain.AI platform, Telegram bot and VKontakte bot, as well as the use of the command “Start the artist” on Sber smart devices, in the Salyut mobile application, also on the ML Space platform.
Recall that the neural network was developed and trained by Sber AI researchers with the joint support of scientists from AIRI Artificial Intelligence Institute on the combined dataset of Sber AI and SberDevices.