In 2023, Sber’s Kandinsky neural network attracted more than 12 million users who generated over 200 million images from text prompts. This surge marks Kandinsky as a standout in user adoption and creative output within the AI image-generation space.
According to Sber’s press release, Kandinsky led in growth rate and ranked just behind Stable Diffusion in popularity among developers, as reported by Hugging Face, a leading source for open source AI solutions. This positions Kandinsky as a major player in the rapidly expanding field of text-to-image models.
Sber explains that Kandinsky can understand requests across a broad set of topics in more than 100 languages. With this neural network, users can produce an unlimited number of photorealistic images in a wide variety of styles, enabling flexible exploration and rapid iteration for projects, education, or personal creativity.
Beyond generation, Kandinsky can modify imagery by replacing individual objects or entire regions, blending multiple images, finishing incomplete visuals, and operating in an infinite canvas mode. The model is attuned to local cultural cues, recognizing architectural landmarks, iconic objects, and folk art elements, which helps generate images that resonate with specific contexts and audiences.
The technology also supports motion, allowing four-second animated clips and full-length videos up to eight seconds in duration, expanding its utility for quick previews and dynamic storytelling. This capability adds a temporal dimension to the creative workflow, enabling creators to convey ideas more vividly.
Kandinsky works within the GigaChat service and integrates with SberBank Online and Salute mobile applications, and it is accessible on Sber smart devices via the voice command Start artist. The model was developed and trained by Sber AI researchers with collaboration from the AIRI Artificial Intelligence Institute, using a combined dataset from Sber AI and SberDevices to enhance its breadth and reliability. The development reflects ongoing investments in multilingual understanding, culture-aware rendering, and user-centric design across Sber’s AI ecosystem, with collaboration from leading researchers and institutions within the company’s network.