Sber presented its own AI video generator based on text description Kandinsky 4.0 Video

At this week’s AI Journey conference, Sber presented a beta version of the Kandinsky 4.0 Video neural network, which allows you to generate realistic videos based on a text description or starting frame. The algorithm can be useful not only for ordinary users, but also for designers, marketers and animators, since it can be used to create various videos: from animated videos with congratulations for loved ones to trailers and clips.

Image source: fusionbrain.ai

The company noted that since the release of the first version of Kandinsky Video last year, the developers have significantly improved many characteristics of the algorithm, including the quality of the videos created and the generation speed. The updated neural network is capable of creating videos up to 12 seconds long with a resolution of 1280 × 720 pixels based on a text description or image. It also supports creating videos with different aspect ratios. The visual quality has improved: the contrast and clarity of frames have increased, the composition has become more precise, and the movements of objects in the frame have become realistic.

In addition to this, the developers announced an accelerated version of the neural network – Kandinsky 4.0 Video Flash, which is capable of generating videos up to 12 seconds long with a resolution of 720 × 480 pixels in just 15 seconds. At the initial stage, representatives of the creative industries, including artists, designers and filmmakers, will have access to the new version of Kandinsky Video through the fusionbrain.ai portal. The algorithm will become available to ordinary users in the first quarter of next year.