Сбер открывает новую главу в истории своих нейросетей, выпуская Kandinsky 5.0 Video Lite. Несмотря на компактный размер в 2 млрд параметров, эта AI-модель обходит по качеству генерации видео предыдущие версии и большинство open-source решений, работая при этом значительно быстрее.
We are releasing Kandinsky 5.0 Video Lite, the first model in the new Kandinsky 5 series. The model runs at 768x512 resolution and, despite its compact size of just 2 billion parameters, demonstrates quality superior to previous versions of Kandinsky and most current open-source state-of-the-art solutions. A key focus is efficiency: the model is compact, requires fewer resources, and generates faster. This result was achieved through a comprehensive approach—from data collection and preparation to pretraining and fine-tuning. We explored modern architecture optimization methods and applied our own developments to balance quality and speed.