Meta has built a web-based detection tool that can identify images and videos created with Muse Image, its new AI image generation model. The tool checks for invisible watermarks embedded by the model, which Meta calls Content Seal. According to Engadget, the watermarks hold up even when an image is cropped, compressed, resized, or screenshotted.
The move comes after Meta’s Oversight Board said earlier this year that it was “concerned” the company was “inconsistently implementing” digital watermarks on AI content created by its own tools. This detection tool looks like a direct response to that pressure. As AI-generated images flood social media, the ability to trace content back to its source is becoming a real issue for platforms and users alike.
Meta’s previous AI image tools added a small visible logo to the bottom right corner of generated images. The new approach drops that visible marker entirely in favor of the invisible Content Seal system. The proprietary watermarking method is a shift from Meta’s earlier open-source versions of similar technology.
The detection tool is straightforward to use. You upload an image, and it tells you whether it carries a Content Seal watermark. A positive result means the image was generated or edited using the Meta AI app or meta.ai. A negative result means it probably wasn’t. In testing, the tool correctly identified both edited images and fully AI-generated ones, including screenshots of those images.
That said, the tool has some clear gaps worth knowing about:
- It only works with images made using Muse Image, not older Meta AI models.
- It is not compatible with SynthID or C2PA Content Credentials, two watermarking standards used by other AI companies.
- The Meta AI app itself cannot perform the same detection check. When asked about an image the web tool flagged as AI-made, the app said it had no ability to verify that.
- The feature has a daily rate limit on identification checks, which feels oddly restrictive for a transparency tool.
The rate limit is a strange design choice. If the goal is to help people verify whether content is AI-generated, capping how many checks a user can run in a day works against that goal. Meta hasn’t explained the reasoning.
Looking ahead, Meta says it plans to extend Content Seal watermarks to AI-generated and edited videos. The company is also working on a separate video generation model called Muse Video, described only as “coming soon.” How well Content Seal holds up across video formats will be a bigger test, since video is harder to watermark reliably and easier to manipulate in ways that might strip embedded data.
The broader context here matters. Across the industry, there is no single agreed standard for AI content detection. Google uses SynthID, the Coalition for Content Provenance and Authenticity backs C2PA, and now Meta has its own proprietary system. The lack of cross-compatibility means a tool built to detect Meta’s watermarks won’t catch content made with Google’s models, and vice versa. For users trying to understand what they’re looking at online, that fragmentation is a real problem that no single company can fix on its own.
