Sber has updated its neural network: with Kandinsky 2.2, you can create photorealistic images with a higher resolution and change the aspect ratio during rendering. This was reported to RB.RU by a company representative.
It is noted that “a significant increase in quality” was achieved when creating portraits. To further train the neural network, a dataset of 1.5 billion text-image pairs was used, which is 300 million more than in the previous version: Kandinsky 2.1
The new version of the model can also generate stickers, from which it will be possible to build sticker packs on Telegram.
In addition, the Kandinsky 2.2 bot has learned to change individual objects or elements in images according to the text description, while maintaining the composition of the original illustration. This was made possible by introducing a special controlled-change building block, ControlNet, into the model.
Kandinsky 2.2 recognizes requests in Russian and English, knows more than 20 drawing styles, can mix various drawings, lay out an image according to the text, generate images similar to the given ones, fill in the missing parts of the image, and create paintings on the infinite canvas mode.
The last time Sberbank updated its neural network was in April, when Kandinsky 2.1 was reported to speed up the creation of high-quality images based on their textual description to a few seconds.
Author:
anastasia mariana
Source: RB
I am a professional journalist and content creator with extensive experience writing for news websites. I currently work as an author at Gadget Onus, where I specialize in covering hot news topics. My written pieces have been published on some of the biggest media outlets around the world, including The Guardian and BBC News.
