Make-a-Video Meta, the company that owns Facebook, has announced a new product that will let users make videos that are made by AI. Make-A-Video, which isn’t out yet but is the latest step toward AI-generated art, will let people make videos based on a text prompt.
The movies are only five seconds long and don’t have any sound, but they show how AI-generated art has changed from still images to short video clips.
The tool was made by a group of engineers at Meta who work on machine learning. They published a white paper on the Cornell University website about their research.
The company has also given movies made with the tool and the text prompts that were used to make them.
“This is a huge step forward.” Video creation is much harder than photo creation because the system must not only correctly create each pixel but also predict how it will change over time. Mark Zuckerberg, who started Facebook and is the CEO of Meta, said:
“Make-a-Video solves this problem by adding a non-supervised learning layer that lets the system senses motion in the real world and uses that information in traditional text-to-image production. We’re looking at this with an eye to the future.” intend to provide a demo. Until then, have fun.
If Meta’s AI model works well, other companies and institutions will probably put more money into AI models that make videos.
Last month, an artist from New York City got the first registered copyright for art made by covertly spreading AI.
Chris Kashtanova got the rights to the graphic novel Zaria of the Dawn, which was made with the commercial AI art generator Mid Trip, according to a statement posted on his Instagram account.
Copyright was checked by UPI using public records in Facebook company
Even though AI-generated art has been registered with the US Copyright Office before, Kashtanova’s claim is the first to use a model based on covert diffusion.
Meta’s model is different from the ones that came before it because it learns from unlabeled video footage instead of human supervision and pairs of images with subtitles.
It makes films by using well-known diffusion techniques, like taking a visual static and showing it until the image in the signal can be seen.
Meta wrote on the Make-A-Video website, “Make-A-Video research builds on recent improvements in text-to-image generation technologies that are meant to allow text-to-video generation.”
“The system learns what the world looks like and how it is usually shown in pictures and words. It also uses videos without labels to learn how the world works. With just a few words or lines of text, Make-A-Video can help you bring your ideas to life by turning them into unique videos.