Google Photos is set to introduce a new feature that will enable users to determine if an image has been modified using artificial intelligence (AI). Images enhanced via the Reimagine tool in Magic Editor will now carry labels created with SynthID, a watermarking technology developed by the tech company based in Mountain View. Users can find this label in the “About this image” section, and it can also be detected through specialized tools. This initiative appears to be a proactive measure to mitigate the prevalence of deepfakes generated through Google’s AI image editing capabilities.
Google Photos to Add AI Watermark
In a recent blog post, Google announced that the AI watermark will be automatically applied to images edited with the Magic Editor specifically using the Reimagine tool, which utilizes the Imagen 3 AI model to introduce new elements into photos.
Moreover, all uses of Magic Editor will result in the application of the watermark. Google has innovated a technology named SynthID, which adds an invisible, imperceptible tag to the image itself.
This tag will be accessible in the About this Image section, and it can also be identified through specific detection tools. Unlike traditional watermarks, this one cannot be removed by cropping, applying filters, or duplicating the image, as it is inherently embedded within the pixels of the image.
Developed by Google DeepMind, SynthID can be integrated into various forms of media, including AI-generated text, images, audio, and videos. The watermarking process ensures that the tag is not merely superficial; for example, in videos, it is embedded within each frame.
The inclusion of watermarks on AI-generated images is aimed at curbing the risks associated with deepfakes—realistic media that has been digitally altered or generated using AI to disseminate misinformation and mislead viewers.
It is important to note that Google has indicated that in cases where the modifications made to an image through Reimagine are minimal, SynthID may not be applied. This could occur in situations where, for example, a user changes the color of a small flower in the background. Users can verify the presence of the AI watermark by examining the metadata of the image.
