Press "Enter" to skip to content

Meta unveils “Make-A-Video”, its artificial intelligence for generating videos

Meta presented, on September 29, his new artificial intelligence (AI) project “Make-A-Video”. The latter allows generate short video from sentence or text. It complements Meta’s research on the development of images through artificial intelligence. For the moment, “Make-A-Video” is in the experimental phase and is unavailable to the general public.

From image generation to video generation

After going out “Make-A-Scene”, an image generation AIlast July, Meta decided to take the difficulty up a notch and tackle video generation. Mark Zuckerberg writes that ” It is much more difficult to generate videos than photosbecause, in addition to correctly generate each pixelthe system must also predict their evolution over time “.

In the same category

Scene where Google Search On 2022 took place.

Google Search On 2022: which ads to choose?

To create these videos of a maximum length of 5 secondsthe AI ​​was trained by a network of 10 million videos. A first database allowed him to know the name of any object while a second taught him how these are supposed to move.

A baby panda swinging

To achieve this result, “Make-A-Video” learned what a panda looks like, how a baby behaves in a swing and how a swing moves in space. Video: Meta.

If it is obvious that these films were made by an artificial intelligence, the first results are promising. Tanmay Gupta, a computer vision researcher at the Allen Institute for AI, remarked to US magazine MIT Technology Review than “Make-A-Video” has a good sense of depth and light. He qualifies all the same by explaining “ that there is still a lot of progress to be madeespecially if these systems are to be used for video editing and professional content creation “.

4 AI-generated videos

Examples of AI-generated videos from a sentence. Video: Meta.

What are the dangers for these generation AIs?

Between Dalle-E and Midjourney, the use of AI to produce images continues to grow in popularity since the beginning of the year. However, as these technologies develop faster and faster, concerns begin to arise. These concerns include the use of “Make-A-Video” for disinformation or propaganda purposes.

Anticipating these questions, researchers at Meta have indicated (PDF) have removed any offensive words or images. A company spokesperson said, ” as part of the search for “Make-A-Video”, we will continue to explore ways to refine and reduce potential risks “.




Spread the love