Google has expanded its AI watermarking technology SynthID to include AI-generated text and video. The announcement was made during the Google I/O 2024 developer conference, where DeepMind CEO Demis Hassabis introduced the enhanced system. SynthID was originally designed for AI-generated images and audio, but now extends its capabilities to additional media formats and improves the identification of AI-generated content.
New features in SynthID
First introduced last August, SynthID initially embedded invisible watermarks into AI-generated images. These watermarks are undetectable to the human eye, but can be identified by specialized systems. Unlike other watermarking protocols such as C2PA that use encrypted metadata, SynthID employs a unique imprinting method. With the latest update, SynthID can now mark AI-generated videos and text, expanding its reach and utility.
How SynthID works
SynthID for text is designed to complement widely available AI text generation models and can be deployed at scale. Adjust the probability scores of tokens produced by a large-scale language model to create patterns that act as watermarks. This method is less effective for responding to factual prompts because there is less opportunity to adjust token distribution without affecting factual accuracy. For videos, SynthID includes every frame in the generated video based on the image and audio watermarking method. This innovative approach embeds invisible watermarks without affecting the quality, accuracy, creativity, or speed of the text or video generation process.
The importance of AI watermarks
With the proliferation of AI technology, the need to watermark AI-generated content is increasing. AI tools have been misused for malicious activities such as spreading political misinformation, fabricating statements, and creating non-consensual explicit content. The enhanced SynthID tool aims to mitigate these risks by providing a reliable way to identify AI-generated material and support efforts to maintain content integrity and authenticity. is.
Microsoft already watermarks images produced by Bing Image Creator. The addition of a watermark is intended to help users identify content generated by her AI. Meta has introduced a mandatory AI watermark across the platform. OpenAI also added watermarking, a way to add metadata to digital content to prove its origin, to strengthen security against deepfakes generated by AI models.
Government and industry safeguards
SynthID is part of a broader effort to develop safeguards against misuse of AI technology. The Biden administration has directed federal agencies to develop guidelines for the use of AI and emphasized the importance of these safeguards. Our work with SynthID aligns with these directives and aims to create a more secure digital environment as AI continues to evolve.
Starting today, all videos generated by Veo on VideoFX will be watermarked by SynthID. Later this summer, Google will open source his SynthID for text watermarks, allowing developers to build with this technology and incorporate it into their models. Additionally, Google plans to publish a detailed research paper on text watermarking technology and update its Responsible Generative AI toolkit to include SynthID.
