Russian ChatGPT – Kandinsky 2.1 starts testing in Kazakhstan
Russian bank and IT giant Sber has begun testing Kandinsky 2.1, a Russian neural network that draws on text queries in Kazakhstan, Turanews.kz reports.
On April 4, a presentation of a generative model based on rudalle took place in Russia. Russian developers have made an analogue of Western applications Midjourney and Stable Diffusion. But, unlike them, the Russian program that draws pictures on text requests for pictures works for free.
The developers offered several options for interacting with Kandinsky 2.1. There is a full version of the application on the Fusion Brain website, where there are about a dozen different styles. In addition, you can generate an image through the telegram bot, where so far there are three styles for creating pictures.
There are also options for mixing two custom images, as well as drawing a picture on a text command.
Kandinsky 2.1 has been trained on one billion text-to-image queries, and now the neural network can draw even the most complex queries.
Experts of the web service for hosting IT projects and their joint development GitHub added the Russian neural network to the TOP-5 most popular networks in the world. Two days after the presentation of Kandinsky 2.1, more than a million unique users were registered, a week later there were twice as many. During this time, more than 10 million images were generated. According to these indicators, the Russian model bypassed ChatGPT-4 from OpenAl.
SberAl developers continue to work on improving Kandinsky 2.1. The service is not planned to be made paid in the near future, unlike its Western counterparts, where a subscription costs $10-$100.