Google Photos' SynthID AI labels on AI content still have a minor flaw

Google is adding AI watermarks to photos edited with its Magic Editor’s generative AI feature, using SynthID, a digital watermarking system created by DeepMind.

The watermark helps identify AI-altered images and is already applied to those generated using Google’s Imagen model.

But still, it comes with a flaw. These SynthID labels on Google Photos may not detect the AI manipulation if it’s “too small,” like, “if you change the color of a small flower in the background of an image.”

With gazillions of AI-generated images flooding the internet, sometimes it’s hard to pinpoint if the tech is behind an image. That can be a problem, especially as AI images are getting more advanced at a time of misinformation like this.

Last year, Meta said that it’s injecting labels to AI-generated images on its social media platforms, including Facebook, Instagram, and Threads. So, you may have seen “Made by AI” or “AI info” labels hanging around when browsing these social media networks.

Though, while its own AI-generated images from Meta AI already have metadata and invisible watermarks, the Facebook parent company developed tools to identify similar markers from other companies like Google, OpenAI, and Adobe. But at that time, it was limited to images, and AI-generated audio and video were largely unmarked.

Both Meta and Google are part of the C2PA initiative that develops open standards for certifying the origin and authenticity of digital content. It provides a framework for embedding tamper-resistant metadata in images, videos, and other media to track their creation, modifications, and provenance.