Google DeepMind, in collaboration with Google Cloud, has recently introduced a groundbreaking tool known as SynthID. This tool is specifically designed for watermarking and identifying AI-generated images produced by Google’s image-generating model.
With SynthID, a digital watermark is embedded directly into the pixels of an image. This watermark is invisible to the human eye but can be easily detected by an algorithm. The tool is currently in beta and is exclusively available for select users of Vertex AI, Google’s platform for constructing AI apps and models. Moreover, it only supports Imagen, Google’s text-to-image model.
The main objective of SynthID is to combat the proliferation of false information and to empower individuals in identifying AI-generated content. Even after modifications like adding filters or altering colors, the watermark remains intact. This is made possible through the utilization of two AI models – one for watermarking and another for identification.
While SynthID may not be entirely foolproof against extreme image manipulations, it presents a promising approach for the responsible utilization of AI-generated content. Currently, there are existing watermarking techniques for generative art. Nevertheless, there is mounting pressure on technology firms to provide clear identification of AI-generated works.
Several other companies, including French startup Imatag and Steg.AI, offer watermarking tools that endure image edits. Furthermore, both China’s Cyberspace Administration and U.S. Senator Kyrsten Sinema have stressed the importance of transparency in generative AI. Notably, Microsoft, Shutterstock, Midjourney, and OpenAI have also committed to watermarking their AI-generated content.
Despite these efforts, a widely accepted standard for watermarking has yet to be established. Presently, SynthID is limited to Imagen. However, there are plans to potentially make it available to third parties in the future. As the demand for AI-generated content continues to grow, it is imperative for tools like SynthID to ensure the integrity and accountability of such content.
