Unlock Editor's Digest free of charge
Roula Khalaf, editor of the FT, picks her favorite stories on this weekly newsletter.
Meta is introducing artificial intelligence models that may generate realistic videos from text instructions that can rival competing offerings for filmmakers and content creators from OpenAI and Runway.
Movie Gen is a set of storytelling models that might be used for a spread of tasks akin to: For example, you possibly can create videos as much as 16 seconds long, edit videos, adjust sounds to videos, and personalize videos with specific images.
The owner of Instagram and Facebook plans to supply video generation tools to Hollywood filmmakers, artists and influencers who create content on its social networks. OpenAI announced its own video generation model Sora in February and introduced it to the film industry, even though it has not yet been released as a product.
While Meta released some examples of videos generated by its models on Friday, it said it doesn't expect to integrate the models into its platforms for users until next yr on the earliest.
“Right now . . . If you were to make use of a video editing or video generation feature in Instagram, it probably wouldn't meet your expectations of the speed that you just want,” said Connor Hayes, vice chairman of generative AI products at Meta. “But Broadly speaking, you may imagine that these models are really powerful for things like Reels creation and Reels editing across the family of apps, and that's the direction we're where we are able to apply them .” Reels is the video creation and sharing feature on Instagram.
The push into video generation is a component of an effort by tech firms to develop tools that might be used more broadly within the entertainment industry, including promoting, as they give the impression of being for tactics to monetize their AI advances. Runway, an AI video generation startup, signed a take care of entertainment company Lionsgate last month to coach a custom model for its library of movies, including and .
Meta claimed its videos outperformed its competitors like Sora and Runway when it comes to “overall quality, movement, naturalness and consistency,” citing blind human reviews.
Its models were trained on “a mix of licensed and publicly available datasets,” Meta said, but declined to offer further details. It has previously used public content from its platforms akin to Facebook and Instagram for its AI.
The realistic nature of AI-generated videos – and the power to recreate human likenesses in them – has raised concerns amongst staff, including actors and production staff, concerning the tools' potential impact on their work in the long run.
“While there are lots of exciting use cases for these basic models, it can be crucial to notice that generative AI shouldn’t be a substitute for the work of artists and animators,” Meta said, emphasizing that it can proceed to hunt feedback from filmmakers and creators.
Meta said it will watermark all videos generated by the model to avoid copyright concerns and problems that might arise with deepfakes. “These are lots of the challenges we must overcome before we are able to responsibly bring a product to market, and are also a very important reason why that is currently a research-only announcement,” Hayes added.
