Google has released a new watermarking and AI picture identification tool. The use of this technique embeds a digital watermark that is invisible to the human eye but can be recognized for verification within an image’s pixels. This program, called “SynthID,” was created in partnership with Google Cloud by Google DeepMind, the company’s AI research group.
Currently, a limited number of users may use SynthID through Vertex AI, Google’s platform for building AI models and apps. It is specifically made to function with Imagen, one of Google’s newest text-to-image models. Text input may be used by Imagen to create incredibly lifelike visuals.
Although generative AI offers enormous creative potential, it also has hazards, including the capacity to inadvertently or purposefully distribute false information. To enable people to recognize when they are interacting with created media and to combat disinformation, Google DeepMind stressed the need of identifying AI-generated material.
Importantly, SynthID keeps working even when the photographs are altered, for as by adding filters, changing the colors, or compressing them.
DeepMind trained two AI models on a variety of photos, one for identification and the other for watermarking, to generate SynthID. It’s crucial to remember that SynthID is unable to positively identify watermarked photos. Instead, it makes a distinction between pictures with watermarks and those without.
The platform views SynthID as a promising first step in allowing the ethical usage of AI-generated material, despite the fact that it is not impervious to severe picture manipulation. As it improves, the tool may handle AI-generated audio, video, and text.
Google hopes to make this technology available to other parties and incorporate it into additional products soon. This endeavor aims accountability and openness in the age of AI-generated information.
Google has released a new watermarking and AI picture identification tool called “SynthID,” created in partnership with Google Cloud by DeepMind’s AI research group. The tool embeds a digital watermark invisible to the human eye but can be recognized for verification within an image’s pixels. Currently, only a limited number of users can use SynthID through Vertex AI, specifically for Imagen, Google’s text-to-image model. DeepMind trained two AI models on various photos, one for identification and the other for watermarking, to generate SynthID. The tool distinguishes between pictures with watermarks and those without. While it is not immune to severe picture manipulation, SynthID is a promising first step in allowing ethical use of AI-generated material. Google plans to make this technology accessible to third-party users and incorporate it into more of its products, aiming to improve accountability and transparency in the age of artificial intelligence-generated information.