sbir updated his generative kandinsky neural network, or rather , opened free access to kandinsky video version 1.1. it was announced a month and a half ago with the previous kandensky update to version 3.1, but see how it is works, it has only become possible now. so, at the output you can get a video with a resolution of up to 512 or 640 pixels over a long country, at 30 frames per second. from the stated changes in comparison with the previous version, the maximum duration of the resulting video has decreased by a quarter to 6 seconds, but the generation process itself has accelerated. in addition, as they say, they doubled the quality of the resulting video, for which... the algorithm worked and reduced the number of artifacts. however, as for the final result, a lot depends on the quality of the product, that is, the text request based on which the video is created. for example, this is how kandinsky’s video imagines a russian processor in a domestic cellular base station, not exactly photorealistic, but very good. but an attempt to create a video about a yandex drone driving pas