Runway launches a new neural network model for creating videos from a text description

Artificial intelligence that converts text into an image is now real mainstream, but similar neural networks for video are only gaining momentum. At the forefront of this direction is the startup Runway, which released already the second AI model that generates short videos based on several words.

Runway offers a web-based video editor that specializes in machine learning tools such as background removal. This company helped develop the Stable Diffusion text-to-image model, and in February announced its first AI video editing model Gen-1.

Gen-1 focused on converting existing footage, allowing users to apply a particular style of overlay from picture to reel. The Gen-2 model is more focused on creating videos from scratch.

The demos shared on Runway are short, unstable, and certainly not photorealistic, but even such frames demonstrate a good understanding of the AI ​​​​of three-dimensional space, proportions and the logic of movement objects.

While Gen-2 is not freely available, a Runway spokesperson assured that the company will "provide wide access in the coming weeks."

The prospects for the introduction of such AI are already scaring experts. After all, neural networks for working with video promise not only new creative possibilities, but also new threats to disinformation.