• By Ashish Singh
  • Mon, 07 Oct 2024 01:04 PM (IST)
  • Source:Reuters

Meta, the owner of Facebook, said that it has developed a new artificial intelligence model called Movie Gen. It claims that this model can compete with tools from top media creation firms like OpenAI and ElevenLabs and can produce realistic-looking video and audio snippets in response to user suggestions.

Meta gave examples of Movie Gen's work that included films of animals surfing and swimming, as well as videos that used people's actual images to represent them painting on a canvas.

In addition, Movie Gen may be used to modify pre-existing videos and produce sound effects and background music that are synchronised with the video material, according to a blog post by Meta. In a couple of these videos, Meta used the tool to alter a parking lot where a man was skateboarding from dry ground to one with a splattering puddle by inserting pom-poms into the hands of the man running alone in the desert.

READ: Meta Fined 91 Million Euros By EU For Storing User Passwords In Plain Text Without Encryption

According to Meta, Movie Gen videos have a maximum length of 16 seconds, while audio files might have a maximum length of 45 seconds. It released statistics from blind testing that demonstrate the model outperforms products from businesses including Runway, OpenAI, ElevenLabs, and Kling.

As it has done with its Llama series of large-language models, Meta representatives stated that the company is unlikely to make Movie Gen available for open use by developers, stating that each model's hazards are evaluated separately.

READ: Meta Connect 2024: Meta AI Brings Real-Time Voice Commands And Celebrity Voices To WhatsApp

Meta is also collaborating closely with the entertainment industry and other content producers on the application of Movie Gen and will integrate it into its own products at some point in the upcoming year. As stated in the tool's blog post and research paper, Movie Gen was constructed by the company using a combination of licensed and publicly available datasets.