Google Introduces SynthID Detector to Combat AI-Generated Images

AI-generated content is becoming increasingly prevalent across the internet. It’s not just limited to social media; you might even find them in printed materials. If you’re someone who usually distinguishes between good and bad imagery, you might catch these AI creations yourself. But, if that’s not your strong suit, Google has released a tool to help you out: the SynthID Detector.

The SynthID Detector derives its name from a digital watermark technology imbued in images produced by Google’s Nano Banana tool within Google Gemini. Essentially, every image generated via Gemini carries an invisible identifier that notes it was created by AI and points back to Google’s platform.

While Google’s SynthID Detector could assist in identifying AI-generated images online, it’s important to mention that it’s not entirely reliable.

Because the SynthID watermark isn’t visible to the naked eye, AI-generated photos can circulate the internet unnoticed. Google aims to change that. The SynthID Detector is designed to identify the origins of certain AI images, though it’s worth stressing that it only applies to a limited subset.

To utilize the SynthID Detector, you need to visit the official website and log in with your Google account. Then, you can drag and drop a suspicious image into the designated search box. If the detection is successful, you’ll quickly see where the image originated from, and it may even indicate if the photo has undergone any edits.

This tool allows for the analysis of a range of file types, including images, videos, and audio files that you might typically encounter online, covering many potential scenarios. However, there are significant limitations you should keep in mind before placing too much faith in it.

Specifically, SynthID Detector is primarily built to recognize images that carry Google’s SynthID watermark, which currently includes images generated via Gemini and a couple of selected partners like Nvidia and OpenAI.

But that’s about it.

Google is working to increase SynthID’s compatibility with other AI image generation entities and services. For example, the Apple Playground app, which is included with newer iPhones, will be supported soon. Yet, a multitude of image creation tools—including Midjourney, Adobe Firefly, and various iterations of Stable Diffusion—will likely evade detection.

In short, while Google’s SynthID Detector may serve a purpose in flagging AI-generated images online, it’s far from perfect. Any images originating from outside of SynthID’s current partnerships will yield no results.

Another limitation is that the SynthID tool doesn’t recognize real images. It mainly searches for SynthID watermarks and cannot identify photos that lack these markers entirely. So, authentic images may be misclassified alongside AI-generated ones from non-partners. It would certainly be a plus if the SynthID Detector could at least affirm the authenticity of real images, but in this regard, it’s not very effective.

As AI continues to gain traction, society may require more resources like SynthID to discern truth from fabrication. While SynthID Detector is a promising starting point, it clearly illustrates the gaps in AI detection capabilities, especially when it comes to the lack of central standards for AI creation and distribution. For detection tools to be truly effective in supporting human needs, they either have to incorporate sophisticated scanning functions or there needs to be standardized procedures in place for AI, neither of which appears to be on the horizon anytime soon.