Google's DeepMind announces SynthID to watermark AI-generated images

Traditional watermarks are layered on top of images, usually in the form of translucent text or design. (Image: Google)

Google's AI unit, DeepMind, has unveiled SynthID, a way to watermark AI-generated images for easier identification.

In a blog post, announcing the new tool, DeepMind said that the technology was developed in conjunction with Google Cloud, and embeds a digital watermark into the pixels of an image.

Also read | Elon Musk reacts to growing list of Indian-origin CEOs in top companies

The watermark will not be perceptible to the human eye, but can be used to detect AI-generated images. For now, the tool is being released in beta and only to a limited number of Vertex AI members, that use Imagen, the company's text-to-image model.

"Being able to identify AI-generated content is critical to empowering people with knowledge of when they’re interacting with generated media, and for helping prevent the spread of misinformation," Google wrote in a blog post.

Traditional watermarks are layered on top of images, usually in the form of translucent text or design. However, these methods have been proven to be insufficient in helping identify AI images.

These watermarks can be easily edited out or cropped out of an image. SynthID has been developed with this in mind. It embeds a watermark directly into the pixels of an image, making it harder to edit out.

Also read | Pixel 8 and Pixel 8 Pro could come with AI camera features: Survey

It is also imperceptible to the human eye, so it doesn't detract from the image. SynthID makes use of two machine learning models, that help it identify and watermark images.

It visually aligns the watermark to the image, and the models help it identify potential AI-generated content. Both models have been trained on a diverse set of images.