Google also has its own Ki system for video creation

Google also has its own Ki system for video creation

San Francisco, Oct. 6, 2022

Google has unveiled its own AI system for creating videos based on text instructions.

The system, called Imagen Video, is based on a „cascade“ of seven video diffusion models, according to Google. The debut comes shortly after the announcement of Meta’s AI-powered online tool Make-a-Video, which also creates short video clips based on text descriptions.

Google Imagen Video is capable of creating 1280×768 videos at 24 frames per second. Videos are currently limited to 5.3 seconds in length. It can create videos in various artistic styles, such as the works of Vincent van Gogh, and create rotating 3D objects while maintaining their structure. Reportedly, the videos still contain obvious artifacts, but have more detail and clarity than previous text video models.

According to Google, Imagen Video was trained using data from the publicly available LAION-400M image-text dataset, as well as 14 million video-text pairs and 60 million image-text pairs. the company has filtered much of the data, but notes that the videos generated may still contain stereotypical, sexually explicit and violent content.