Meta tests new tool to detect AI-generated images from Muse Image

Meta is previewing a web-based detection tool that identifies images and videos created with its newest AI model, Muse Image. Karissa Bell reports for Engadget that the tool checks for invisible watermarks called Content Seal, which Meta says remain intact even after cropping, compressing, resizing or screenshotting an image.

Content Seal marks a shift from Meta’s earlier approach, which added a visible logo to AI-generated images. The version used in Muse Image is proprietary, though Meta has previously released open-source versions of similar watermarking technology, the company tells Engadget. The tool currently only detects content made with Muse Image, but Meta plans to extend Content Seal to AI-generated and edited videos, including output from its upcoming Muse Video model.

According to Meta’s FAQ, a positive result means an image was generated or edited using the Meta AI app or meta.ai. A negative result means it likely was not.

Limitations remain

Bell’s testing revealed several gaps. The detection tool is not yet built into the Meta AI app itself: when asked directly, the app’s assistant said it had no way to verify whether an image was AI-generated. The tool also failed to identify images made with older versions of Meta’s AI models, and it is not compatible with other watermarking standards like SynthID or C2PA Content Credentials. Users also face daily rate limits on how many images they can check.

The rollout comes after Meta’s Oversight Board criticized the company earlier this year for what it called inconsistent implementation of watermarks on its own AI-generated content.