Meta Movie Gen: Zuckerberg creates surprisingly realistic AI videos.
Movie Gen outperforms other comparable models in the industry in these tasks when evaluated by individuals.
Meta has recently launched its new generative artificial intelligence system for videos, known as Meta Movie Gen. This innovative tool allows for the creation of realistic-looking video and audio clips. The company showcased several 10-second examples generated with Movie Gen, including scenes like a baby hippo styled after Moo Deng swimming, highlighting its advanced capabilities. According to Meta, the latest research reveals that it is possible to use simple text inputs to create personalized videos and sounds, edit existing clips, and transform images into unique videos. Furthermore, it is claimed that Movie Gen outperforms other similar models in the industry based on human evaluations.
Movie Gen features four main functionalities: video generation, personalized video generation, precise clip editing, and audio generation. This system has been trained using a combination of authorized and publicly accessible datasets. Beyond directly generating text-to-video clips, this tool also allows for specific edits in existing videos, such as adding objects to characters' hands or modifying the appearance of surfaces. For example, in one of the demonstration videos, a woman wearing a virtual reality headset was transformed to appear as if she were using steampunk-style binoculars.
Alongside Movie Gen, Meta also introduced Meta Audio Gen, a model that can generate high-fidelity audio of up to 45 seconds from videos and optional text messages. This system includes ambient sound, sound effects, and background music that syncs with the video content.
The company has indicated that it will continue to work with filmmakers and creators to refine these models, seeking feedback to create tools that enhance user creativity in new ways.
In the realm of video editing, Adobe is also integrating generative artificial intelligence capabilities into its Premiere Pro suite, expanding its Adobe Firefly model. Meanwhile, OpenAI has launched Sora, an AI model that impressively showcases videos entirely generated by artificial intelligence, yielding highly realistic results.
At the same time, Meta is exploring generative AI in the film domain with Emu Video, which will enable more professional production of clips that can be easily shared on social platforms like Instagram and Facebook. This approach aims to facilitate editing tools that allow users to modify images and generate engaging visual content directly through their applications.
