Meta quietly launched Content Seal in July, its homegrown invisible watermarking system for AI-generated images. On the surface, it sounds a lot like Google’s SynthID — it embeds a hidden provenance signal that survives cropping, resizing, and screenshots. But the similarities end when you look at execution. SynthID is already baked into Google’s Gemini chatbot and has been adopted by OpenAI. Content Seal, meanwhile, exists as a standalone web tool with a daily rate limit. You can’t even use it inside Meta’s own AI chatbot.
I tested it. An image generated by Meta’s Muse model passed right through Gemini’s detector and the C2PA portal without raising a flag. That means Content Seal watermarks are invisible to every detection system except Meta’s own, which is a problem if the goal is actually helping people spot deepfakes across platforms. Meta spokesperson Faith Eischen told The Verge the company is “exploring ways to bring detection closer to where people encounter AI-generated content” and is “determined to work with our industry peers.” Both phrases suggest the real infrastructure isn’t built yet.
Why reinvent the wheel? Meta sits on the C2PA steering committee alongside Google. It already has a seat at the table where cross-industry standards get hammered out. Instead, it shipped a proprietary system with fewer features than what Google offers — no video support yet, no in-chatbot detection, and a rate-limited checker that feels designed more to protect the technology than to scale transparency. Google and OpenAI impose similar limits, but C2PA doesn’t, and that feels like a missed opportunity to actually lead.
I’m not convinced Meta has thought through the endgame here. If Content Seal only works inside Meta’s walled garden, it’s not a detection standard — it’s a compliance checkbox. The Oversight Board asked Meta to use its own tools to fight deceptive AI content. Shipping a watermark that nobody else can read isn’t quite what they had in mind.
