Google DeepMind Introduces SynthID, A Watermarking Tool for AI-Generated Images
The AI branch of Google, in collaboration with Google Cloud, has developed a groundbreaking tool named SynthID to identify and watermark AI-generated images.
SynthID embeds a digital watermark directly into the pixels of an image. This renders the watermark invisible to the human eye but detectable by an algorithm.
The watermarking technique has been particularly developed for images created using Imagen, Google's text-to-image model, exclusively available within the Vertex AI framework.
SynthID is currently in its beta mode and is accessible to a group of users of Vertex AI. This is a platform of Google dedicated to developing AI models and applications.
Synthid Reflects Google's Commitment To Classify AI-Generated ContentWith the popularity of AI generative models increasing, concerns have surfaced about the misinformation and misuse of AI-generated images. With Google introducing SynthID, the move shows Google's dedication to addressing these concerns.
While generative AI can unlock enormous creative potential, it also introduces new risks, such as enabling creators to spread misinformation - intentionally or unintentionally.DeepMindDeepMind believes that the power to classify AI-generated content would help individuals distinguish between genuine and generated media, which can curb the spread of false information.
SynthID is capable of detecting even modified images. DeepMind stated that the tool can withstand edits like filter application, changes in color, and heavy compression.
The process largely depends on two AI models the company trained simultaneously on a diverse set of images. While one was trained for recognition, the other was trained for watermarking.
However, the tool cannot warrant 100% accuracy in identifying watermarked images. It is capable of distinguishing between images that are likely to be AI-generated and the ones that are genuine.
Other Watermarking Tools Are Quickly Catching UpWhile SynthID has come to the limelight for its watermarking features, similar techniques have emerged from other quarters, too. A French startup called Imatag, founded in 2020, has developed a watermarking tool resilient to cropping, editing, resizing, and compression.
Similarly, there's a tool called Steg.AI that leverages AI models to apply watermarks capable of withstanding different forms of editing.
Leading companies like Shutterstock, Microsoft, and OpenAI have expressed their commitment to watermarking AI-generated images.The mission to ensure accountability and transparency in AI-generated content has also spurred action in the tech industry. China's Cyberspace Administration recently introduced regulations that mandate AI vendors to label generated content.
On the other hand, Senator Kyrsten Sinema in the US stressed the importance of maintaining transparency in generative AI during a committee hearing of the Senate.
The development of SynthID further demonstrates Google's commitment to maintaining this transparency. However, this tool wouldn't be applicable globally.
Presently, SynthID only supports Imagen-generated images. However, DeepMind is exploring the possibility of making it available to third parties in the future.
The post Google DeepMind Introduces SynthID, A Watermarking Tool for AI-Generated Images appeared first on The Tech Report.