Sber Launches Kandinsky 6 Image: Image Generator Gets Faster and Learns to Restore Photos

Sber has updated its model for working with images and integrated Kandinsky 6 Image into GigaChat. The new version focuses not only on text-to-image generation, but also on full-fledged editing: from restoring old photos to precise retouching of individual details and preparing images for further animation.

What Changed

Kandinsky 6 Image is available directly within GigaChat, and for users this is perhaps the main practical change. The model can be used for free and without limits on the number of tasks, which means this is not a one-time demonstration of capabilities, but a service designed for regular use. According to Sber, the new version has become twice as fast as the previous one, handles complex requests better, and produces more detailed images.

This is important not only for those who like to experiment with prompts. The faster the model responds and the more accurately it preserves details, the closer it comes to being an everyday tool for designers, marketers, small shop owners, and content creators. In such scenarios, what matters is not generation as an impressive function, but a predictable result that can be used in a post, product card, banner, or presentation without extensive rework.

From a technical perspective, the model has acquired a Mixture of Experts architecture. This is an approach in which not the entire neural network as a whole is engaged for each request, but only the necessary fragments. For the user, this translates not into abstract metrics, but into faster processing and fewer errors in complex scenarios where many conditions need to be maintained simultaneously: style, composition, characters, clothing, background, and small objects in the frame.

New Scenarios

The main difference between Kandinsky 6 Image and a typical image generator is an expanded set of editing tools. Now the model addresses more tasks that previously had to be solved in several applications or manually, especially if the user needs not to invent an image from scratch, but to carefully improve an existing frame. The focus here is clearly on everyday scenarios, not just on demonstrating technological capabilities in promotional videos for regular users and teams.

• Restoration of old photographs
• Styling of images under different visual formats
• Neural photoshoots based on user images
• Changing clothing and background in photos
• Retouching and virtual makeup

Sber separately emphasizes that local edits have become more precise. This is an important point for everyone who works not on generation from scratch, but on editing an already completed image. In such tasks, users typically need to change one part of an image without destroying everything else. If the model truly preserves the original scene better and makes changes to individual elements more carefully, this makes it significantly more useful for everyday and commercial use.

Search and Animation

Another notable innovation is built-in image search. If the model doesn't know what a character or object that the user is asking about looks like, it can independently find a visual reference. In practice, this should reduce the number of misses in situations where textual description is insufficient or where an up-to-date reference is important, rather than an averaged interpretation from the training data.

For services of this class, this is a way to reduce the number of invented details and bring the result closer to user expectations. Generated images can be immediately animated through the linked Kandinsky Video model, which already works in the GigaChat ecosystem. This turns the service from a simple image generator into a broader visual constructor: first the user gets a static frame, and then can quickly turn it into a short moving clip without switching to another product.

For Sber, this is another step toward GigaChat being perceived as a universal AI assistant, not just as a text chatbot.

What This Means

Kandinsky 6 Image shows that competition in generative graphics is shifting from the question "who draws best from prompts" to a more applied scenario: who will give users a complete set of tools in one window. For the market, this means an increase in the number of mass-market services where generation, retouching, reference search, and basic animation are assembled into a single workflow without unnecessary friction and switching between different applications for a mass audience.