Instagram is working on a new artificial intelligence (AI) feature that will allow users to re-imagine their videos. The AI-powered video editing tool is built on Meta’s Movie Gen AI model, an early research project that can generate video from text prompts. On Instagram, instead of letting users create videos from scratch, the feature will add enhancements to videos shot using a camera and featuring real humans. Users will be able to make changes to the video such as changing the outfit, background and even their overall look.
The new feature was teased by Instagram head Adam Mosseri in a Reel. In a minute-long video, he demonstrated some of the capabilities of the new AI feature and also explained how it would work. Specifically, an AI video editing tool is under development, and Mosseri highlighted that it could be shipped next year.
In the video, the tool can change Mosseri’s outfit, add a gold chain around her neck, and even add a hippopotamus playing in the swimming pool in the background. The tool was also shown to combine some of these effects to make the Instagram head appear to be sitting near a snowy mountain and in a desert landscape.
Captioning the video, Mosseri revealed that the AI video editing tool is powered by Meta’s Movie Gen AI model that was previewed in October. Notably, it is a multi-modal model that comes with four capabilities – Video Generation, Personalized Video Generation, Precise Video Editing, and Audio Generation. It can also combine video and audio generation to create videos with synced audio. It is not currently available in the public domain.
“I’m very excited about Movie Gen, our early AI research model that will let you change almost any aspect of your video with a simple text prompt. Hoping to make it to Instagram next year,” Mosseri wrote in the caption.
The under-development AI feature appears to simply borrow the precise video editing capabilities from Movie Gen. Meta described in the announcement post that the model can make localized edits such as adding, removing or changing elements, and make global changes such as background or style modifications. The AI model also preserves the original content and targets only relevant pixels.
