Meta has published research on a new generative artificial intelligence foundation model called ´Movie Gen´ that can generate and edit short videos and produce synchronized audio from text prompts. The company said the work builds on earlier efforts such as ´Make-A-Scene´ and ´Llama Image´ and combines their modalities to give creators finer-grained control. For now, Meta is sharing the model with a limited set of employees and outside partners while it evaluates a path to product integration next year.
Technically, ´Movie Gen´ can produce 16-second clips at 16 frames per second while accounting for object motion, subject-object interactions and camera motion. It can create personalized clips from a subject´s image plus a text prompt, and it can edit existing footage by adding, removing or replacing elements and changing background or style. The model is designed to target only relevant pixels, preserving remaining original content. On the audio side, Meta says the model can generate up to 45 seconds of ambient sound, effects and music that sync to visual action and narrative cues.
Meta is approaching release cautiously. Bloomberg reported that the company plans to hold back public access until safeguards are in place to stop the technology from being used to create videos of people without consent. In addition, Meta previously said it will label manipulated content on three of its platforms as ´Made with AI´. The company also emphasizes collaboration with filmmakers and creators to refine the tools and avoid replacing the work of artists and animators.
Potential use cases range from short-form social videos on Reels to personalized animated greetings on WhatsApp, and from simple edits to more ambitious creative workflows. The announcement underscores ongoing tensions between innovation, creator rights and deepfake risks. Meta´s next steps will focus on safety, labeling and creator feedback as it tests ´Movie Gen´ with partners and prepares selective integration into its app ecosystem next year.
