The business shared several 10-second videos generated with Movie Gen, including a Moo Deng-esque child crocodile swimming around, to show its abilities. This Movie Gen news comes shortly after its Meta Connect occurrence, which showcased fresh and refreshed technology and the most recent version of its massive language design, Llama 3. 2. 2. Although the tool is not yet available for use, this announcement comes shortly after its Meta Connect event.
Beyond producing easy text-to-video clips, the Movie Gen design can make specific adjustments to an existing clip, such as adding an object to a person’s hand or altering a surface’s appearance. In one of the case videos from Meta, a lady who was wearing a Multimedia headset transformed into a woman with a steampunk binoculars appearance.
With Movie Gen, sound bites can be created to go along with the video. An AI person stands near a fountain in the trial clips, where loud splashes and the ominous sounds of a music are heard, a sports car’s engine purrs and tires slam as it zips around the track, and a snake slides along the forest floor while accompanied by dramatic horns.
In a research paper released on Friday, Meta added more information about Movie Gen. Movie Gen Video consists of 30 billion parameters, while Movie Gen Audio consists of 13 billion parameters. ( A model’s parameter count roughly corresponds to how capable it is, by contrast, the largest variant of Llama 3.1 has 405 billion parameters. ) High-definition videos up to 16 seconds can be produced using Movie Gen, which claims to be better than competing ones in terms of overall video quality.
CEO Mark Zuckerberg demonstrated the Imagine Me feature of Meta AI earlier this year by posting an AI image of himself drowning in gold chains on Threads. Users can upload a photo of themselves and role-play their faces in various scenarios. With the Movie Gen model, it’s possible to create a video version of a similar feature. Think of it like ElfYourself on steroids.
What background has Movie Gen received training in? The specifics are unclear in Meta’s announcement post, which states that” We’ve trained these models on a combination of licensed and publicly available data sets.” For generative AI tools, the sources of training data and what’s fair scrape from the web are still contentious, and it’s rare to find out what text, video, or audio clips were used to create any of the main models.
It will be interesting to see how long it will take Meta to make Movie Gen readily available. The announcement blog feigns to mention a “potential future release.” In contrast, OpenAI released its AI video model, Sora, earlier this year, but it has n’t yet made it public or announced a date of release ( though WIRED did receive a few exclusive Sora clips from the company for an investigation into bias ).
Considering Meta’s legacy as a social media company, it’s possible that tools powered by Movie Gen will start popping up, eventually, inside of Facebook, Instagram, and WhatsApp. Google, a rival in September, announced plans to include elements of its Veo video model for creators to use in its YouTube Shorts sometime in the future.
While larger tech companies are still hesitant to make fully available video models to the general public, you can still use AI video tools from upcoming, smaller startups like Runway and Pika. Give Pikaffects a shot if you’ve ever wondered what it would be like to have yourself cartoonishly crushed by a hydraulic press or completely melt in a puddle.
