Film · Model Watch
Meta's new Muse model can edit video in DaVinci Resolve—and run on your own machines
Muse Spark Open brings an unusually powerful AI coding model to developer-controlled hardware, with a demonstration of how an agent can examine footage and help shape an edit.
Reporting updated Oct 19, 2026
THE TAKEAWAY
Meta has released Muse Spark Open, a powerful open-weight AI model designed to write code, use tools and carry out complicated tasks. Among its demonstrations is something directly relevant to filmmakers: an agent working inside DaVinci Resolve, reviewing footage frame by frame, finding promising moments and reframing them for vertical video. The bigger opportunity is an AI production assistant that creators can run and customize themselves.
WHAT HAPPENED
Imagine giving an assistant a folder of footage and asking it to find the moments that would make the strongest short-form video.
It watches the material, evaluates individual frames, identifies useful moments and helps prepare a vertical edit.
That's one of the workflows Meta tested with Muse Spark Open.
Released October 15, the model can process text, images, audio and video. It is designed for long-running tasks involving software tools, code and multiple decisions.
Meta also demonstrated Muse Code using the model to fix bugs, add features and complete a build of Ice Cubes, an open-source iOS application.
Those examples reveal the direction Meta is exploring: agents that can operate production software rather than merely explain how to use it.
The release arrives just days after Meta introduced Muse Wonder and Muse Wonder Full, larger closed models aimed at sophisticated coding and personal-agent tasks. Muse Wonder is accessible through Meta's developer API and Muse Code; Wonder Full is available through Muse Code subscriptions.
Muse Spark Open takes a different route by giving developers access to model weights they can download, run and customize under Meta's license.
WHY IT MATTERS
AI video generation receives enormous attention, but filmmaking includes plenty of work that happens after images exist.
Footage needs to be reviewed. Moments need to be selected. Clips need reframing, organizing and finishing.
Those jobs require an understanding of the material as well as the ability to operate editing tools.
Muse Spark Open brings those capabilities into the same kind of agentic workflow already reshaping software development.
For independent filmmakers, a future assistant that understands a project's footage and can help execute editing tasks could save significant time, particularly when preparing multiple versions for different platforms.
The practical limits matter. The desktop configuration Meta demonstrated is expensive, and the DaVinci experiment is not the same as a finished consumer editing plugin. Developers can also use hosted versions rather than buy the hardware.
Open weights offer more control over deployment and customization, although the Muse Model License includes commercial conditions.
WHAT CREATORS SHOULD KNOW
Video understanding
The model accepts footage as input, helping agents inspect visual material and make decisions about an edit.
Software control
Meta demonstrated an agent carrying out work in DaVinci Resolve, including identifying strong moments and adjusting vertical framing.
Coding and tools
Muse Spark Open is built to work through programming tasks, call tools, inspect results and correct failures.
Local operation
The model can run without a hosted inference service, but Meta's showcase setup used two Apple M5 Ultra Mac Studios with a combined terabyte of memory.
The interesting step is moving from an AI that suggests an edit to one that can help carry it out.
VEYR ANGLE
There's a useful distinction between generating a beautiful shot and completing a piece of filmmaking.
The first produces material. The second involves selection, timing, judgment and a great deal of production work.
Muse Spark Open is interesting because Meta is exploring that second challenge.
It won't replace an editor's sense of story. But an assistant that understands footage, handles repetitive tasks and works inside familiar software could give small creative teams more time for the decisions that actually make a film worth watching.
