Meta Unveils Movie Gen: Crafting the Future of AI-Generated Videos and Audio Clips

To look at this article again, go to My Profile and then click on Saved stories to see it.

Meta Unveils Movie Gen, Producing Authentic AI-Generated Videos

Meta has recently launched a new AI tool named Movie Gen, designed to create lifelike video and audio content.

The business released several brief videos created using Movie Gen, showcasing a baby hippo reminiscent of Moo Deng as it swims, to highlight what the tool can do. Although the tool isn't ready for public use, the reveal of Movie Gen followed closely on the heels of its Meta Connect event. There, they introduced new and updated devices as well as the most recent update to their expansive language model, Llama 3.2.

The Movie Gen model advances past merely creating basic text-to-video sequences. It has the capability to perform precise modifications to already existing footage, such as inserting an item into a person's grip or altering how a surface looks. In a demonstration video by Meta, they showcased this ability by altering footage of a woman with a VR headset to make it appear as if she was donning steampunk-style goggles.

A video created by artificial intelligence, produced in response to the request "transform me into an artist."

A video created by artificial intelligence depicts a female DJ playing music on turntables. She is dressed in a pink coat and oversized headphones, with a cheetah positioned beside her.

Movie Gen enables the creation of sound clips in addition to videos. In the demonstration videos, there's an AI figure positioned close to a waterfall, where you can hear the water crashing down and the optimistic tones of an orchestra; the sound of a sports car's engine humming and the squeal of its tires as it races around a circuit, and a snake moving through the jungle underbrush, with dramatic horn music adding to the tension.

On Friday, Meta unveiled additional information regarding its Movie Gen project through a published research document. The Movie Gen Video segment is built on 30 billion parameters, and the Movie Gen Audio segment encompasses 13 billion parameters. (The number of parameters in a model is generally indicative of its potential capabilities; for context, the most extensive model of Llama 3.1 is designed with 405 billion parameters.) Movie Gen is capable of generating high-definition videos with a duration of up to 16 seconds, and according to Meta, it surpasses rival models in terms of the quality of the videos produced.

At the beginning of the year, Mark Zuckerberg, the CEO, showcased a new function from Meta AI named Imagine Me. This innovative feature allows individuals to insert their own photo and simulate various situations with their face. Zuckerberg illustrated its capabilities by sharing an AI-generated image of himself covered in gold chains on Threads. Moreover, there's a potential for a video adaptation of this concept through the Movie Gen model, essentially an enhanced version of the ElfYourself application.

Meta's announcement does not specify the exact data Movie Gen has been trained on, merely stating that it involves a mix of licensed and publicly accessible data sets. The origin of training materials and the debate over what constitutes fair use of internet content continue to be hot topics in the realm of generative AI. The specifics regarding the text, video, or audio clips utilized in the development of leading models are often kept under wraps.

Observing the timeframe for Meta to widely release Movie Gen will be intriguing. The company's announcement post ambiguously hints at a possible release down the line. In contrast, OpenAI revealed its own AI video model, Sora, earlier in the year but has still not opened it up to the general public or announced any specific launch dates (although WIRED was given several exclusive Sora clips by the company for a bias inquiry).

Given Meta's background in social networking, it's likely that features driven by Movie Gen may eventually be integrated into platforms like Facebook, Instagram, and WhatsApp. In September, rival company Google announced intentions to offer components of its Veo video model to creators on its YouTube Shorts platform in the coming year.

Major tech corporations are currently hesitant to launch their video-based AI models to the general public. However, you can explore AI video applications today through emerging startups such as Runway and Pika. Try out Pikaffects for a fun experience if you've ever wondered about looking like you're comically squashed by a hydraulic press or instantly dissolving into a puddle.

You May Also Be Interested In…

Direct to your email: A selection of the most fascinating and peculiar tales from the archives of WIRED

Elon Musk poses a threat to national security.

Conversation: Meredith Whittaker Aims to Disprove Capitalist Ideals

What's the solution to a challenge such as Polestar?

Occasion: Be part of The Major Interview happening on December 3rd in San Francisco.

Additional Content from WIRED

Critiques and Tutorials

© 2024 Condé Nast. All rights reserved. Purchases made through our website may result in WIRED receiving a share of the sales, as a result of our affiliate agreements with various retailers. Any content found on this site is protected under copyright and cannot be duplicated, shared, broadcast, stored in any form, or utilized in any way without explicit written consent from Condé Nast. Choices regarding advertisements are available.

Choose a global website