Stable Video Propagation: The company that creates visuals with artificial intelligence now also produces videos

Platform Stable Diffusion, one of the most popular language models for generating AI-based illustrations, has announced an evolution. Now it can also create videos.

Presented this Tuesday (21), Stability’s new feature is called Stable Video Diffusion and is the company’s first generative model for animated clips.results that are based on existing tools and look impressive.

Below is a preview of what the Stable Video Delivery video model can do:

According to Stability, the idea is to use this technology in areas such as advertising, education and entertainment, among many other sectors.

How Does Stable Diffusion Video AI Work?

The company’s generative video AI has the ability to create two- or three-dimensional animations in a few seconds from user text commands. It follows the same logic as other similar services, providing more complex and accurate results based on the amount of detail in the prompt.

It uses the same technical foundations as a static renderer to create illustrations and create animated content. 14 to 25 different frames with motion speed up to 30 frames per second.

In addition to text input, it also has the ability to convert static images into animation, that is, to “bring life” to photographs or drawings.

According to Stability, the generative AI was trained with “millions of videos collected from across the internet” and improved on a smaller sample of around a million clips. However, the company has not fully disclosed whether this set contains copyrighted materials.

Moreover, there are still limitations to the creation in terms of framing, camera effects, duration, and the representation of human faces, at least for now.

How to use Stable Propagation video AI?

Stable Video Diffusion released for now in preview only for researchers in the field. Anyone who wants to use the platform must sign a commitment to the rules of the tool, including not representing real people in productions.

The code is available on the GitHub repository and more technical details can be read in the academic paper published by the team (in English).