Source: Sber press service
Sber presented an updated version of the Kandinsky 2.2 neural network. It generates photorealistic images in the best quality, does an excellent job with portraits, and can draw in over 20 styles.
The neural network was developed and trained by Sber AI researchers with the support of scientists from the AIRI Institute of Artificial Intelligence. The previous model, Kandinsky 2.1, was presented by Sber in April of this year. It has become the fastest growing neural network in the world, gaining one million users in just 4 days after launch.
Source: Sber press service
Kandinsky 2.2. accepts applications in Russian and English. Creates images with higher resolution and allows you to change the aspect ratio during rendering. The neural network can mix multiple drawings, design an image according to a text description, generate images similar to the given ones, fill in the missing parts of the image, and create paintings in infinite canvas mode.
Source: Sber press service
Another innovation: Kandinsky 2.2 can change individual objects or elements in images based on the text description, but at the same time preserve the composition of the original illustration. To do this, the developers introduced a special building block of managed changes to the model – ControlNet. And the updated version can also generate stickers, from which you can collect complete sticker packs.
Source: Sber press service
– We keep improving our neural network to create unique content. The new version gives users even more options while generating better quality images. The neural network is no longer just trying to imitate human creativity, but can also create new meanings and artistic interpretations, says Alexander Vedyakhin, first vice-president of Sberbank’s board of directors.
Source: Sber press service
Kandinsky 2.2 is in the public domain. You can try it for free on the model’s promotion page, on the FusionBrain.AI platform, on the Telegram bot and the VKontakte bot, as well as using the “Start Artist” command on Sber smart devices, on the Salut mobile application. The model is available on the ML Space platform in the DataHub pre-trained models and datasets hub.