Today, hardly anyone remains indifferent to artificial intelligence. “The creative abilities of neural networks have tremendous potential. They are by no means competing with humans; on the contrary, they open up new possibilities for creativity and artistry. For designers, generative models have already become an essential tool for increasing efficiency and breaking free from routine,” says our guest Denis Dimitrov, the Executive Director of Data Research at Sberbank’s AI Future Technologies Department.
At our request, Denis conducted a webinar titled “Russian Innovation Night: Kandinsky and GigaChat – Multimedia Neural Networks for Generating Images and Texts” at the platform of our Russian House. You can watch the recording of the broadcast here.
This article will provide a brief summary of the meeting.
Currently, there is Kandinsky 2.2. The first version was launched in the spring of 2023, and over this short period, the network has undergone significant changes, noticeably improving its generation level. The main feature of the new version is called photorealism.
The Kandinsky 2.2 neural network, like its competitors, creates images based on textual requests. One of the main bonuses is support for over 100 languages, including Russian.
The main modes of operation of the neural network include creating or, in user terms, generating an image based on a textual request; the ability to mix images; creating a new stylistic image based on an existing one; adding details left out of the frame; transferring the outline of an image to a newly generated one.
In the near future, an animation creation service is expected. It is currently in test mode.
The next meeting with Dimitrov was devoted to a master class, and it will be discussed in another blog.
All the images in this blog are made with Kandinsky 2.2 in the theme of “Golden Autumn” but in different styles.