Google Pictures is including its digital SynthID watermarks to pictures which were edited utilizing the Magic Editor’s generative AI characteristic. The brand new characteristic is rolling out “this week” based on Google, and is meant to make it simpler for individuals to shortly establish photos which were manipulated utilizing the “reimagine” software in Magic Editor.
SynthID is a watermarking system created by Google’s DeepMind workforce that embeds a digital metadata tag instantly into photos, video, audio, and textual content to establish in the event that they have been created or altered utilizing AI instruments. The watermark is already being utilized to photographs which are completely AI-generated utilizing Google’s Imagen text-to-image mannequin. Related watermarking methods have additionally been developed by different corporations, just like the Content material Credentials that Adobe applies to works created or edited utilizing its suite of Inventive Cloud apps.
That’s an issue the brand new watermarking replace is making an attempt to unravel, however there are just a few points. For one, SynthID doesn’t visibly alter the picture it’s connected to and could be recognized utilizing a devoted AI detection software that’s part of Google’s “About this picture” characteristic. Google additionally says that some edits made utilizing the Magic Editor reimagine “could also be too small for SynthID to label and detect.”
Whereas watermarking tech may also help if it does catch manipulations, some specialists on this area agree that watermarking alone gained’t be efficient sufficient to reliably authenticate AI-generated content material at scale, and would require a assorted suite of various approaches working in tandem.
