Google's New Site Checks If Images Are AI Fakes

Google just launched a new website called SynthID.com that lets you upload an image or video to see if it was made with AI. This is a big deal because fake AI images are everywhere, and it's getting harder to tell what's real. The tool looks for invisible watermarks that Google and some other companies add to AI-generated content. But here's the catch: it only works for images made with tools from Google, OpenAI, Nvidia, and Kakao. If someone used a different AI tool, like Grok or Midjourney, SynthID.com won't detect it. So you might think an image is real when it's actually fake.

We tested it with a fake image of Ken Paxton, Donald Trump, and Grimace on a golf course. Gemini, Google's AI, correctly said it was AI-generated but then made up a story about the original photo. When we tried a Grok-made image, SynthID.com said it was 'unlikely' made with AI, which is misleading. So the tool isn't perfect. It can give false negatives and even false information. That means you shouldn't rely on it alone to verify images.

Another issue: you need to log in with a Google, Apple, or ChatGPT account to use SynthID.com. That's a small hurdle, but it might stop some people from using it. Also, the tool only tells you if a watermark is present. If it's not, the image could still be AI-generated by a non-partner tool. So it's not a definitive answer.

Overall, SynthID.com is a useful first step, but it's not a silver bullet. As AI images get better, we need more cooperation between companies to make detection reliable. For now, be skeptical of what you see online and use multiple methods to check.