Meta is working on an ID photo and video tool created with its new image generation model, Muse Image. The company showed a preview of a web-based tool that can check the invisible watermarks used by the new model.
This watermarking system, called Content Seal, remains “even if cropped, compressed, resized, or screenshotted,” Meta explained in a blog post. “We’re previewing a detection tool that allows you to check if an image carries the Content Seal watermark, giving you the first step to help you better understand if an image was created with Meta AI.”
Content Seal appears to be a new form of Meta. The partial version of Muse Image is proprietary, although the company has previously released open source versions of the technology, Meta told Engadget. The new Meta models do not include any visible watermarks, as previous versions of Meta AI added a small logo in the lower right corner.
Currently, Meta AI’s detection capabilities are limited to images created or edited with Muse Image, although the company said it plans to expand Content Seal watermarks and AI-generated videos. Meta is also working on a different video production model called Muse Video “coming soon.”
I tried the new detection feature on the photos I created today with Meta AI and the web-based tool was able to find the watermark of edited photos and completely AI creations (like the one shown above). It also got the watermark on the screenshots of my photos. “A positive result means that the image was generated or edited using the Meta AI application or meta.ai,” the company explains in the FAQ. “A negative result means that the image is less likely to be processed using the Meta AI application or meta.ai.”
Interestingly, the new detection capabilities of Meta AI do not appear as part of the Meta AI app yet. When I asked Meta’s app-based assistant about the image the web tool identified as AI-generated, it replied that it had no testing capabilities. “I can’t tell you for sure that this particular image is made of Meta Al just by looking at it,” he said. “Meta Al doesn’t automatically watermark itself, and I don’t have a tool that can tell which model of Al makes up the image.”
Meta has previously faced criticism for how it labels and identifies AI-generated content in its apps. The Oversight Board told the company earlier this year that it was “concerned” that Meta was “inconsistently using” digital watermarks in AI content created by its tools.
The new feature still seems to have some limitations, though. Content Seal is not compatible with SynthID or C2PA Content credentials, two standard marking methods used by other companies. The web-based feature was unable to identify images created or edited with earlier versions of Meta’s AI models in my testing. When I added images created in old chats with Meta AI, it couldn’t tell me that the image was created with its AI. The feature also appears, for some reason, to be below the limits of the Meta rating. After uploading a few examples, I was notified that I had reached my “daily limit of diagnostic tests.”
