SynthID embeds subtle changes into individual pixels of images, making the watermarks invisible to the human eye but detectable by computers.
Google’s DeepMind, the company’s AI research arm, is testing SynthID, a digital watermarking technology aimed at identifying images created by artificial intelligence (AI) to combat the spread of disinformation.
SynthID embeds subtle changes into individual pixels of images, making the watermarks invisible to the human eye but detectable by computers. While it’s acknowledged that the system isn’t foolproof against extreme image manipulation, it represents a step forward in tackling the challenge of differentiating between real and AI-generated images.
As AI-generated images become more prevalent, including popular tools like Midjourney with over 14.5 million users, concerns about copyright and ownership have arisen. Google’s own image generator, Imagen, will be subject to the watermarking system.
Unlike conventional watermarks, which can be easily edited out, SynthID’s watermarks remain detectable even after manipulation. Google’s Pushmeet Kohli stated that even if an image is cropped, resized, or its colors altered, the watermark will still be discernible by DeepMind’s software.
This experimental launch of SynthID is in line with Google’s commitment to ensuring safe AI development.
However, DeepMind acknowledges the need for more user interaction to refine the system’s robustness. While companies like Microsoft and Amazon have also pledged to watermark AI-generated content, standardization among different watermarking methods remains an ongoing debate in the industry.
Beyond images, Meta, formerly Facebook, has indicated that its unreleased video generator, Make-A-Video, will incorporate watermarks to address demands for transparency over AI-generated video content.
Conversely, China has taken a more aggressive approach by banning AI-generated images without watermarks entirely this year. Companies like Alibaba have adhered to this directive, applying watermarks to images created using their text-to-image tool to ensure transparency and authenticity.
Google DeepMind, also known as DeepMind Technologies Limited, is a British-American artificial intelligence research facility. It was established in 2010 and was purchased by Google in 2014.
As a result, it is now a fully owned subsidiary of Alphabet Inc. Following a corporate reorganization in 2015, Google is now headquartered in London and has research facilities in Canada, France, and the United States.