Chinese developers have created a neural network called CogVideo that can generate short videos from text descriptions. CogVideo can create videos with a relatively high frame rate – 4 frames per 4 seconds.

Currently, the neural network is “lion drinking water”, “woman riding a horse in the sea”, “man eating pizza”, etc. can create videos for As for the principle of operation, it is similar to that of the DALL-E 2 and Imagine neural networks, which creates images based on textual description.

Even the previous version of CogVideo is claimed to outperform all common neural network models in terms of job quality. So far, the authors have not provided any source code for the program or models to learn their neural networks.

Earlier, scientists said that the neural network is in the process of creating its own language, and experts have not been able to decipher it. Source