Google DeepMind has unveiled SynthID, a watermarking/identification tool for generative art aimed at enhancing transparency in AI-generated images.
Indeed, recently, Generative art or images created using AI tools have the potential to produce deepfakes and ethical implications in the art community. With the help of this watermarking tool, it can assist in identifying content produced by AI.
This technology embeds a digital watermark that is invisible to the human eye directly into the image pixels. Initially, SynthID will be applied to a number of clients using Imagen, Google’s artificial art generator.
SynthID incorporates two learning models: one for watermarking and one for identification. Both models result in optimized machine learning models designed to recognize watermarked content and enhance the imperceptibility of the watermark on original content.
While not perfect, watermarking is seen as a promising technical solution to enable responsible use of AI-generated content. Google hopes that this tool can be extended to other AI models, including text, video, and audio.
Although it can help address deepfake issues, the implementation of this digital watermarking will undoubtedly face opposition from parties disagreeing with its application. To strengthen this purpose, Google strives to make SynthID available to third parties, even for free, to enhance AI transparency across the industry.”
