Meta is working on a tool to identify images and videos created with its new image generation model, Muse Image. The company previewed the web tool that can check for invisible watermarks used by the new model.
This watermark system, called Content Seal, stays in place “even when cropped, compressed, resized or captured,” Meta explains in a blog post. “We’re previewing a detection tool that lets you check if an image has a Content Seal watermark, providing a first way to help you better understand if an image was created with Meta AI.”
Content Seal seems to be a somewhat new approach for Meta. The version that is part of Muse Image is proprietary, although the company has previously released open source versions of the technology, Meta told Woozad. Meta’s new models don’t include any visible watermarks, like some previous versions of Meta AI that added a small logo in the bottom right corner.
For now, Meta AI’s detection capabilities are limited to images created or edited with Muse Image, although the company has announced plans to expand Content Seal watermarks to AI-generated and edited videos as well. Meta is also working on a separate video generation model called Muse Video that will be “coming soon.”
I tried the new detection feature on images I created today with Meta AI and the web tool was able to detect a watermark for both edited images and completely AI creations (like the one pictured above). It also found the watermark in screenshots of my images. “A positive result means that the image was generated or modified using the Meta AI application or meta.ai,” the company explains in an FAQ. “A negative result means the image is unlikely to have been processed using the Meta AI application or meta.ai.”
Interestingly, Meta AI’s new detection capabilities don’t appear to be part of the Meta AI app yet. When I asked Meta’s app-based assistant about an image the web tool had identified as AI-created, it told me it didn’t have the ability to verify it. “I can’t tell you for sure if this specific image was made with Meta Al just by looking at it,” he said. “Meta Al does not automatically watermark images and I do not have a tool that can detect which Al template created an existing image.”
Meta has previously faced criticism over how it labels and identifies AI-generated material in its applications. The Oversight Board told the company earlier this year that it was “concerned” that Meta was “inconsistently implementing” digital watermarks on AI content created by its own tools.
The new feature appears to have other limitations, however. Content Seal is not compatible with SynthID or C2PA Content Credentials, two established watermarking methods used by other companies. The web 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 older threads with Meta AI, it couldn’t tell me if the image was created with its AI. The feature also seems, for some reason, to be subject to Meta’s rate limits. After downloading a handful of samples, I was alerted that I had reached my “daily identity checks limit.”
