Google DeepMind and Google Cloud have unveiled a watermarking tool that adds a digital marker to an image generated using artificial intelligence (AI).
SynthID will initially only be available to a limited number of Vertex AI customers using Google’s Imagen text-to-image model.
However, Google said it could be expanded for use across other AI models and will be integrated into more of its own products, with plans to offer it to third parties in the near future.
“While generative AI can unlock huge creative potential, it also presents new risks, like enabling creators to spread false information — both intentionally or unintentionally,” Google said.
“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.
“This technology embeds a digital watermark directly into the pixels of an image, making it imperceptible to the human eye, but detectable for identification.”
Google explained that SynthID used two deep learning models — for watermarking and identifying — that were trained together on a diverse set of images.
“The combined model is optimised on a range of objectives, including correctly identifying watermarked content and improving imperceptibility by visually aligning the watermark to the original content.”
Google said using a traditional watermark to identify AI images was insufficient, as they could easily be edited out.
“For example, discrete watermarks found in the corner of an image can be cropped out with basic editing techniques,” the company said.
It also said that it was important to strike a balance between imperceptibility and robustness of the watermark.
“Highly visible watermarks, often added as a layer with a name or logo across the top of an image, also present aesthetic challenges for creative or commercial purposes,” the company explained.
“Likewise, some previously developed imperceptible watermarks can be lost through simple editing techniques like resizing.”
Google also said that metadata information could be manually removed or lost when files are edited.
SynthID will remain detectable even after modifications like filters, colour changes, and compression.
Google acknowledged although the technology was not perfect, its internal testing showed it could withstand many common image manipulations, as illustrated below.