Meta introduces an AI detection tool for images and videos created with Muse Image

Meta has introduced a new detection tool designed to identify images created or edited using its latest artificial intelligence image generation model, Muse Image. The web-based feature can detect invisible digital watermarks embedded in AI-generated content, allowing users to determine whether an image was produced with Meta AI.

The company said the new technology is based on a watermarking system called Content Seal. Unlike visible logos or labels, the watermark is hidden within the image and is designed to remain detectable even after common edits such as cropping, resizing, compressing or taking screenshots. Meta said the feature is intended to improve transparency around AI-generated content and help users understand the origins of digital images.

Invisible watermark aims to improve AI transparency

Meta explained that Content Seal has been developed to withstand many of the changes people typically make before sharing images online. According to the company, the invisible watermark remains intact “even when cropped, compressed, resized, or screenshotted.” The company added, “We’re previewing a detection tool that lets you check whether an image carries a Content Seal watermark, providing an initial way to help you better understand if an image was made with Meta AI.”

The detection tool currently works only with images generated or edited using Muse Image through the Meta AI app or the Meta AI website. If the tool detects the watermark, users receive confirmation that the content was produced or modified using Meta AI. If no watermark is found, Meta says it is unlikely the image was generated with its current AI tools.

Unlike some earlier versions of Meta AI, Muse Image does not place a visible watermark or logo in the corner of generated images. Instead, the company has moved entirely to the hidden Content Seal approach. Meta noted that while it has previously released open-source watermarking technologies, the version used in Muse Image is proprietary and has been developed specifically for its latest AI models.

The company also confirmed that the technology will not remain limited to still images. It plans to expand Content Seal to cover AI-generated and AI-edited video content as well. This move is expected to coincide with the release of Muse Video, Meta’s upcoming AI video generation model, which the company said will arrive soon.

Early testing shows strengths and limitations

Initial testing of the detection tool suggests it successfully identifies both fully AI-generated images and images edited with Muse Image. It also appears capable of recognising the watermark even when images are captured via screenshots, demonstrating the resilience of the embedded digital marker.

Meta explains the results in a frequently asked questions document. “A positive result means that the image was generated or edited using the Meta AI app or meta.ai,” the company said. “A negative result means it is unlikely that the image was processed using the Meta AI app or meta.ai.”

However, the detection capability is currently limited to the standalone web tool. The Meta AI assistant itself cannot yet determine whether an existing image was created using the company’s AI technology. When asked about an image that the web-based detector had identified as AI-generated, the assistant responded, “I can’t tell you definitively if this specific image was made with Meta AI just by looking at it. Meta AI doesn’t automatically watermark images, and I don’t have a tool that can detect which AI model made an existing image.”

The separate availability of the web tool suggests Meta has not yet integrated the identification system across all of its AI products. Users must therefore upload images to the dedicated website to verify whether Content Seal is present.

The company has also introduced usage restrictions for the service’s preview version. After a limited number of uploads, users may receive a notification stating that they have reached their daily identification limit, preventing further checks until the limit resets.

Meta continues work on AI content identification

The introduction of Content Seal comes as technology companies face increasing pressure to identify AI-generated media clearly. Concerns about misinformation and manipulated content have led many firms to develop watermarking systems to distinguish AI-generated content from human-produced content.

Earlier this year, Meta faced criticism over its handling of AI-generated content. The company’s Oversight Board expressed concern that Meta was “inconsistently implementing” digital watermarks on content created by its own AI tools. The latest detection system appears to be part of Meta’s broader effort to improve transparency and address those concerns.

Despite its advances, the new system still has several limitations. Content Seal is not compatible with other widely used industry standards, including SynthID and C2PA Content Credentials, which several other technology companies support. This means the detection tool cannot verify AI content that uses those alternative watermarking methods.

The system also cannot identify images created using older versions of Meta’s AI image generators. Images produced in previous Meta AI conversations without the new Content Seal watermark are not recognised by the detection tool, limiting its ability to identify historical AI-generated content.

As Meta continues expanding its AI capabilities, the company is expected to enhance both its content generation models and its detection technology. Extending invisible watermarking to AI-generated video could be an important step toward helping users distinguish synthetic media as AI tools become more widely adopted.