Google Creates 'Imperceptible' Watermark for AI-Generated Images

pcmag.com:

Google is showing off a system that can hide a watermark in AI-generated images without changing the how the pictures look. 

The company’s “SynthID” system can embed digital watermarks in AI images that are “imperceptible to the human eye, but detectable for identification,” Google’s DeepMind lab says.

Google isn't disclosing how SynthID creates these imperceptible watermarks, likely to avoid tipping off bad actors. For now, DeepMind merely says the watermark is “embedded in the pixels of an image,” which suggests the company is adding a small, minute pattern alongside the pixels that won’t disturb the overall look. 

The company creates the watermarks using two deep learning models that are trained to improve the system’s imperceptibility while still correctly identifying the digital watermarks.   

DeepMind added: “We designed SynthID so it doesn't compromise image quality, and allows the watermark to remain detectable, even after modifications like adding filters, changing colors, and saving with various lossy compression schemes—most commonly used for JPEGs.” The watermark can also remain in the image even if it’s cropped. 

The company added: “SynthID isn’t foolproof against extreme image manipulations, but it does provide a promising technical approach for empowering people and organizations to work with AI-generated content responsibly.” 

Google is launching SynthID as a beta for select customers of Imagen, the company’s text-to-image generator available on the Vertex AI platform. The system can both add the watermark to an image and also identify pictures that carry the digital stamp. 

Google says it could expand the system to other AI models, including its own products. The tech giant also hopes to make SynthID