Meta presented an AI content detector for images and videos generated by Muse Image

Meta demonstrated a preliminary version of a web tool for detecting images and videos created using the Muse Image generator: the application detects invisible watermarks that the model automatically embeds in each generated file.

How the detector works

The principle is simple. A user uploads an image or video to a web application, and it analyzes the file for invisible digital marking that Muse Image automatically embeds during generation. If the watermark is detected — the content is identified as AI-generated.

Invisible watermarks are one of the key technical approaches to marking AI content. Unlike obvious logos or text labels, they do not change the visible appearance of the file, but are preserved during standard manipulations — compression, resizing, publication through social media. That is why the approach is attractive: users get "clean" content, while platforms and regulators retain the ability to verify its origin.

The reliability of the system depends on the resistance of the marking to external impacts. Specialized tools to remove watermarks can reduce detection effectiveness. Meta has not yet disclosed details about the durability of its solution.

• Muse Image embeds an invisible watermark in each generated file
• The detector checks images and videos for the presence of this marking
• The tool is presented as a web application in preview status
• Meta has not announced a public release date

Why does Meta need a detector of its own AI?

Developing a verifier for its own generative product is not a paradox but a logical consequence of regulatory pressure. Meta manages the world's largest photo and video platforms — Facebook and Instagram — and finds itself at the center of discussions about the traceability of synthetic media.

European AI Act requires marking synthetic content. In the United States, similar bills are being promoted. In this regulatory context, creating a detector is not an act of goodwill, but a basic requirement for a platform of such scale.

A number of AI companies already support the C2PA standard (Coalition for Content Provenance and Authenticity), which embeds metadata about origin directly into the file. Meta with invisible watermarks moves in the same direction: it makes verification part of the infrastructure, not an optional add-on.

What is Muse Image

Muse Image is a generator of images and videos from Meta. A distinguishing feature: automatic marking is built into the model from the start, not added afterward. Each file from Muse Image already contains a watermark — and now Meta has shown a tool that can read this mark.

The link "generator + detector from one vendor" signals that content traceability has become a product priority, not a reaction to pressure.

What this means

If Meta brings the tool to public release, it will create one of the first widely accessible verifiers of AI content from a platform of such scale. For newsrooms, fact-checkers, and moderation services, this potentially simplifies the identification of synthetic media. The key condition: the system works only as long as generators consistently embed marking, and users do not have simple tools to remove it.

*Meta is recognized as an extremist organization and is banned in the Russian Federation.