(New York) Meta presented on Friday its new generative artificial intelligence (AI) interface Movie Gen, which allows you to create sound videos from text or images, as other AI pioneers already offer. .
Movie Gen is Meta’s third image generation model after a first version in July 2022 and a second in November 2023.
None of the three interfaces have been made publicly available. They are tested at this stage by creators and directors who evaluate Movie Gen and help improve the software.
Asked by AFP regarding the date of putting Movie Gen online, Meta did not immediately respond. In addition to demonstration videos, the group published a research paper on Friday describing its model.
Language models are software based on huge databases, which allow them to generate content, text, image, video, sound or computer code in response to a request formulated in everyday language.
Movie Gen can produce a video from a written request, but also from a simple photo. It also offers editing functionality for an existing video.
The model also offers to add a soundtrack to the video, on the same principle of request in everyday language.
In the short sequence published on the Meta site, one sentence is enough to dress up the video of a quad driving in the desert with a roaring engine sound and a guitar tune.
Meta suggests that a version of Movie Gen could one day be usable via its social networks Instagram and Facebook, or even through WhatsApp messaging.
The group takes the example of a video which would be edited using instructions in everyday language before being published on social networks.
Meta is not the first AI player to develop a video generation model.
Runway, from start-up Runway AI allows, from a written request, to create a shot of a few seconds, to transform a series of still images into a short video, or to reshape an existing sequence to transform, for example, a photo into a painting .
In February, OpenAI also launched its version, called Sora, while Google is working on a model called Lumiere.
Meta claims that Movie Gen “is superior to similar industry models when its output is evaluated by humans.”