Google is adding new digital watermarks to photos that have been edited using the company’s Magic Editor. This generative AI feature debuted with the Google Pixel 8 series and is included in Google’s other recent Pixel smartphones.
Magic Editor enables photographers to tap on subjects or objects in their photos to move (or remove) them and uses AI to make wholesale changes to the scene, including adjusting the background and lighting conditions.
There are notable concerns about misinformation using generative AI technologies, which Google partially addressed by implementing SynthID digital watermarking on images created entirely by the company’s AI. However, this SynthID digital watermark now applies to images edited using Magic Editor’s “Reimagine” feature, at least in most cases.
Google notes that in some cases, users’ edits with Reimagine “may be too small for SynthID to label and detect.” Google offers an example of someone changing the color of a small flower in the background of a shot as one that may evade detection. However, generally speaking, the company believes its SynthID changes will ensure that people can better identify AI-generated content “quickly and easily.”
SynthID has been in development since 2023 and embeds a digital watermark directly into the pixels of an image. While this is not visible to people, it is easily detected by compatible software and platforms, including Google Image search.
The move also addresses some serious concerns about Google’s lack of AI transparency. When the Pixel 9 Pro launched, PetaPixel reported that there was no watermark or information in an image’s “info” tab about whether it had been edited using the company’s generative AI. The potential problems are obvious, including quickly creating and spreading disinformation.
“Images made from scratch using Gemini’s image generator use a tagging system called SynthID which tags them as not real, but images altered using generative AI in the Pixel 9’s Magic Editor do not. It’s a frankly shocking, woeful oversight,” wrote PetaPixel‘s Jaron Schneider last August.
Now, images edited using Magic Editor will (usually) have SynthID digital watermarking and information about AI editing will be included in an image’s information.
However, there remain concerns about easily identifying AI-edited content on the web. A digital watermark like this requires that someone consume content within a compatible ecosystem, which is not always the case. Identifying AI-altered content requires a multi-faceted approach, and while Google has improved one of those facets today, there are still gaps left to close.
Image credits: Google
Source link
