Google’s SynthID Detector Is Now Open to Everyone to Spot AI-Made Images

Summary

Google moved its SynthID detector from limited access to a global English-language release. It checks images and video for invisible watermarks, and partners now include OpenAI, Nvidia, Kakao, and Apple. It only catches content from tools that add the watermark, and Google says Apple Intelligence images will carry it soon.

Why it matters

Anyone can now check whether a suspicious photo came from an AI tool that uses the watermark. It is a handy check, but a clean result does not prove an image is real.

HTD Says

A free way to check if that viral photo is AI-made is useful, as long as you remember it only catches the tools that play along. AI detection will become increasingly important and prevalent. And while, as people know, it can’t check everything, it can probably help flag some generated images from big-name AI services.

It’s unknown whether you can run an AI-created image through a filter to remove the watermark or ID. But I’m guessing people will reverse-engineer SynthID (probably using AI) to create “detector cleaners.” It’s a whole new market that is coming online.

First we had AI content creation, then AI detector services (think about that college paper that is AI-generated), then countermeasures for those AI detector services, and then embedded text watermarking from Anthropic and OpenAI. The same process will happen with images.

Source:

9to5Google