Sber has unveiled its latest neural network, Kandinsky 6.0, which allows users to produce video clips complete with synchronized speech, music, and ambient sounds in Full HD quality. This user-friendly tool requires no special expertise to operate.
Key features of the new software include the ability to create videos lasting up to five seconds with accompanying audio, lip-sync functionality ensuring that characters’ mouths move in time with the dialogue, and sound quality that meets the streaming standard at 44 kHz. Additionally, it offers realistic physics for movement and supports various resolutions, including SD, HD, and Full HD.
In terms of performance, Kandinsky 6.0 shows significant improvements, being 71% more effective than its predecessor, Kandinsky 5.0. It also surpasses Veo 3.1 Fast in image animation and outperforms LTX 2.5 in various metrics.
For developers interested in utilizing this technology, Kandinsky 6.0 is available as open-source software under the MIT license. It can be accessed through platforms such as FAL, Diffusers, FastVideo, ComfyUI, SGLang, and vLLM-omni. Anton Frolov from Sber stated, "Anyone can build their own services on top of it—free of charge and without restrictions."
Informational material. 18+.
