Meta's Movie Gen brings AI video and audio together

Anyone who has experimented with AI video knows the frustration: the clips look stunning, but they are completely silent, and changing a single detail usually means throwing the whole prompt away and rolling the dice again.

Meta just unveiled Movie Gen, an ambitious research project that tackles video generation, precise editing, personalized character preservation, and audio sync under a single umbrella.

At the center of this release are two large models. Movie Gen Video is a 30-billion-parameter transformer model that can produce high-definition videos across different aspect ratios directly from a text prompt. Meanwhile, Movie Gen Audio is a 13-billion-parameter transformer model dedicated to sound. It takes video input alongside optional text instructions to generate synchronized ambient audio, foley effects, and instrumental music with high fidelity.

Frame-accurate edits and personalized avatars

Beyond basic text-to-video prompts, Meta is focusing heavily on post-production control.

Movie Gen allows creators to take an existing or generated clip and perform precise edits through simple text instructions. You can add, delete, or swap out specific objects in the frame without breaking the rest of the scene, or apply global changes like altering the background lighting and visual style.

Personalization is another major piece. By uploading a single photo of a person alongside a text prompt, Movie Gen can generate dynamic videos that keep the subject's identity and facial structure intact while delivering natural motion.

To see how the tech holds up in Hollywood, Meta partnered with horror powerhouse Blumhouse. A group of selected filmmakers tested the suite before its public debut, including director Aneesh Chaganty, who created a short project pointedly titled "i h8 ai" to explore what these tools can actually do.

Why it matters

We think Meta is making a very calculated move here by shifting the AI video race from raw generation to realistic editing workflows.

Right now, using AI video in production is a disjointed mess. You generate silent footage in one tool, find sound effects in another, and pray you do not spot an artifact that forces a complete re-render. By bundling text-based in-painting, character consistency, and synchronized sound design into one stack, Meta is aiming directly at the daily headaches of creators.

If this reaches creators in a usable form, it could seriously streamline pre-visualization, concept testing, and quick turnaround content creation for solo animators and video editors alike.

However, there is a catch. Movie Gen is currently just a research preview with a "potential release" planned down the road. Running a 30B video model paired with a 13B audio model requires massive compute. Until Meta proves it can run this affordably at scale without stripping away its best features, Movie Gen remains an impressive research milestone rather than an immediate everyday tool.