Google will put invisible watermarks on AI-generated images
%2Favatars%2FEwuSxftyWaVt.png&w=96&q=75)
%2F2023%2F08%2F30%2Fimage%2Fjpeg%2FOUpPbytf2tHxTuVDS8BzOv4YkAN1BBT5R0TaWbM8.jpg&w=3840&q=75)
There are often tell-tale signs when an image is generated with the help of artificial intelligence. Occasionally, the creator includes it in the image's description, while in other cases, one must watch for irregularities such as absent features, unusual green flowers, or oddly smoothed backgrounds.
However, there are synthetically produced images that are truly indistinguishable from reality. AI detection tools like Maybe's AI Art Detector are available to determine the extent to which AI has contributed to an image's creation.
Joining the league of AI detector tools, Google DeepMind, the tech behemoth’s AI arm, unveiled SynthID on Tuesday, which works on a two-pronged technology. This tool is capable of watermarking AI-generated images and can also recognize them.
No degradation of image quality
Based on two deep learning models, for watermarking and identifying, both have been trained together on a diverse set of images. Watermarks on images have long been used by photographers and online visual media companies to prevent copyright infringement. These watermarks can be easily edited out of the image. But SynthID’s watermarks aren’t run-of-the-mill.
The watermarks are granular and embedded into the pixels of the image. One cannot differentiate between the watermarked and non-watermarked images because they are not visible to the naked eye but can be detected using SynthID. One can add filters, change color composition or do lossy compressions, but it won’t affect or water down the watermark.
SynthID is currently only being released as a beta version to its Vertex AI customers who are using Google’s text-to-image AI generator called Imagen. Vertex AI is a unified AI platform by Google that offers all of Google's cloud services under one roof.
“While generative AI can unlock huge creative potential, it also presents new risks, like enabling creators to spread false information — both intentionally or unintentionally. 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,” said the blog.
Full launch at a later date
Google DeepMind has categorically warned that SynthID isn’t a foolproof tool against extreme image manipulations, but it does empower people and organizations to work with AI-generated content responsibly.
The blog doesn’t mention if SynthID, as a standalone tool, possesses the capability to identify AI-generated images which do not have its digital watermark. However, it can identify AI-generated images which have been created using Imagen and if the creator has applied the digital watermark.
Google DeepMind has also announced that it plans on scaling up the technology to be used in identifying AI-generated text, audio, and video.
Will Google deploy SynthID at a larger scale?
Google Images search offers access to more than 136 billion pictures. Earlier this year, the company introduced a novel function named 'About this image' that supplements images with extra information. This information encompasses the image's initial indexing date in Google's database and the website where the image was originally published. Google wants to help users identify the original source of the image. This will help in weeding out the photorealistic AI generated images.
With deep fakes and tools that enable one to create deep fakes on the rise, there were images of Pope Francis wearing a huge Balenciaga jacket floating on the internet. There was also one of Elon Musk kissing a robot. They both looked realistic. Of course, they were both figments of someone’s imagination and a result of using a deep fake tool.
It would be interesting to see how Google applies its latest AI image detector SynthID and other similar applications to segregate fake images.
SHOW COMMENT (1)
/2023/09/23/image/jpeg/9fsKSWRxyPmZZ6TbJoC0urwAKJUdeMfRlPRYW3CR.jpg "Earth’s crust and mantle grew in sync, study finds")
/2023/09/23/image/jpeg/cGOeGwdWJ02fZSawTOjXC9mbyIH0ihwm8Mnpn7bJ.jpg "ISRO tries to wake up Chandrayaan-3 after lunar night, no response yet")
/2023/07/17/image/jpeg/45219tvYuQJId0ojEgO8xNJ447jHljZ4msSiFffI.jpg "Manganese ocean ‘potatos’ are highly radioactive")
/2023/08/17/image/jpeg/wkfQOEAHMgv7YekMUPWWlPJGY7Fn1Pa2blepLs4m.jpg "Too hot to fly? The effects of heatwaves on air travel")
/2023/07/29/image/jpeg/7tlA3v28MfILdeoVd77ZS8anzOgVIc22cYqsIxdt.jpg "How high heat affects EVs and what you can do about it")
/2023/09/23/image/jpeg/EkaUs1PFMs22T2bo39a8T3BslgZ8qxot4rm0p91i.jpg "Upcoming Northern Lights to be the strongest in 20 years")
/2023/09/23/image/jpeg/Kap6NAa8IoTyQmZQEwB4waL2MWykBpyJc9piLpRK.jpg "The genius engineering behind 3D-printed rockets")
/2023/09/12/image/jpeg/E1LfUWjqRrg0bFqBnRzB7XzmfJAd6YPNclMl5I9n.jpg "How AI can revolutionize health on long-duration spaceflights")
/2023/08/27/image/jpeg/fi494iLoRqITLtvrXotofLEJk8JZH1s8Z3Upge1M.jpg "Gallant ships: 5 underdogs of the seas that punched above their weight")
/2023/09/04/image/jpeg/UEuPO8vP8YA3PICuhwuvmebK4Pl9Cm1cShTAqGB2.jpg "Chinese researchers create dancing microrobots using lasers")