Meta previews Content Seal watermarking for Muse Image

In short: Meta has previewed Content Seal, an invisible watermark for images made with its Muse Image model, plus a separate web tool to check for it.

What happened

Meta says every image generated or edited with its Muse Image tool now includes Content Seal, an invisible watermark. An invisible watermark is like a hidden stamp inside the file that people cannot see, but a detector can look for.

Meta also previewed a web based detector that scans images for this hidden stamp. Meta says the stamp should still be detectable after common changes like resizing, compression (saving as a smaller file), and some cropping.

Independent testing has raised questions about how well it works in real use. Reuters tested 40 AI generated images and the detector confirmed all of the originals, but it failed to verify 55 percent after the images were cropped down to about one third to one half of their original size. Meta told Reuters the tool is a preview and that heavy cropping can remove the signal.

Content Seal is also limited in where it works. It currently only detects images made or edited with Muse Image, and it is not built to recognize other watermark systems such as Google DeepMind’s SynthID, or broader standards like C2PA Content Credentials (a shared “nutrition label” style record for digital media).

Why it matters

Watermarks are one of the main ways companies are trying to help people tell whether an image was made by AI. If a system only works inside one company’s products, or if it breaks after common edits like cropping, it can be harder for regular people, journalists, and platforms to use it as a reliable check.

Source: The Verge AI