Sber has released Kandinsky 6.0 Image Pro.

Sber has released Kandinsky 6.0 Image Pro. The main update here is in editing. In a side-by-side model, it compares with Flux 2 Max and GPT Image 1.5, evaluating specific aspects: how editing integrates into the scene, whether style, geometry, face, background, and local consistency are preserved. This is usually where image editing models fail. There is also an increase in speed: more than 40% compared to the previous version due to MoE, parallelization of inference, and attention optimization. The most interesting part is Image RAG. The model pulls relevant images into context and hits specific details more accurately, which in regular generation can easily turn into 'something approximately Slavic.' Use cases include object and text removal, item replacement while preserving style, stylization while maintaining the face, restoration of archival photos, interiors by plan, and facades with one prompt.