Kandinsky 2.1, the free image-generation model from Sber, is rapidly expanding as a major AI service in the global landscape. Sber’s communication team highlighted the pace at which the platform is gaining traction, underlining its position as one of the fastest-growing AI offerings available today.
Developers note that Kandinsky 2.1 achieved a striking milestone in just four days by attracting its first one million unique users. That speed surpasses the early growth curve of other leading models, including OpenAI’s ChatGPT, which reached the same threshold in five days. This rapid uptake signals both broad curiosity and practical utility across diverse creative workflows.
Since the launch, the platform has supported the creation of more than 10 million images, with user counts climbing to over two million. The service demonstrates a high level of accessibility and versatility, enabling a wide audience to explore visual ideas with ease and speed. The system is designed to respond to textual prompts in natural language and deliver results in seconds, turning imaginative concepts into visuals with impressive responsiveness.
Kandinsky 2.1 supports multilingual interaction, covering 101 languages, and opens up opportunities for creative collaboration across borders. Its capabilities include blending multiple sketches into cohesive compositions, refining or completing drawings, and operating in an endless canvas mode that allows interior and exterior artwork to evolve in real time. The model represents a collaborative effort by Sber AI researchers, with joint contributions from the AIRI Artificial Intelligence Institute, leveraging a combined dataset assembled by Sber AI and SberDevices to broaden its expressive range.
The technology builds on a robust foundation from its predecessor, absorbing weights learned from a vast training corpus. The latest model incorporates a large-scale training regime that includes one billion text-image pairs from prior work, along with 170 million high-resolution pairs. It then further refined its capabilities using an independently curated dataset consisting of two million high-quality image pairs, enabling more precise alignment between text prompts and visual outcomes.