Identifying AI-generated content with SynthID
We’re beta launching SynthID, a tool for watermarking and identifying AI-generated content. With this tool, users can embed a digital watermark directly into AI-generated images or audio they create. This watermark is imperceptible to humans, but detectable for identification.
Being able to identify AI-generated content is critical to promoting trust in information. While not a silver bullet for addressing the problem of misinformation, SynthID is an early and promising technical solution to this pressing AI safety issue.
This technology was developed by Google DeepMind and refined in partnership with Google Research. SynthID could be further expanded for use across other AI models and we plan to integrate it into more products in the near future, empowering people and organizations to responsibly work with AI-generated content.
How does SynthID work?
SynthID uses two deep learning models — one for watermarking and another for identifying:
- Watermarking: SynthID uses an embedded watermarking technology that adds a digital watermark directly into AI-generated content. The combined model is optimized to improve imperceptibility by aligning the watermark to the original content.
- Identification: SynthID can scan the image or audio for its digital watermark and help users assess whether the content, or part of it, was generated using our AI models.
SynthID for AI-generated music
In November 2023, SynthID was expanded to watermark and identify AI-generated music and audio. SynthID's first deployment will be through Lyria, our most advanced AI music generation model to date. SynthID will watermark any content published by our Lyria model.
SynthID’s watermark is embedded directly into the audio waveform of AI-generated audio.
First, SynthID converts the audio wave, a one dimensional representation of sound into a spectrogram. A spectrogram is a two dimensional visualisation that shows how the spectrum of frequencies in a sound evolves over time. Once the spectrogram is computed, the digital watermark is added into it. Finally, the spectrogram is converted back to the waveform. During this conversion step, SynthID leverages audio properties to ensure that the watermark is inaudible to the human ear so that it doesn’t compromise the listening experience.
The watermark is robust to many common modifications such as noise additions, MP3 compression, or speeding up and slowing down the track. SynthID can scan the audio track to detect the presence of the watermark at different points to help determine if parts of it may have been generated by Lyria.
SynthID for AI-generated images
To help identify AI-generated images, SynthID is available to a limited number of Vertex AI customers using the Imagen suite of our latest text-to-image models that use input text to create photorealistic images.
SynthID adds a digital watermark directly into the pixels of AI-generated images, making it imperceptible to the human eye.
We designed SynthID so it doesn’t compromise image quality, and allows the watermark to remain detectable, even after modifications like adding filters, changing colours, and saving with various lossy compression schemes — most commonly used for JPEGs.
SynthID can scan the image for a digital watermark and provides three confidence levels for interpreting the results for identification. If a digital watermark is detected, part of the image is likely generated by Imagen.
Note: The model used to produce synthetic images and audio on this page may be different from the model used on YouTube, Imagen and Vertex AI.