SynthID
In plain wordsGoogle's system for hiding an invisible watermark in pictures, video, audio and text made by its AI tools, so they can later be recognized as AI-generated.
A family of AI watermarkingAI watermarking, n.: Hiding an invisible signal in AI-generated text, images, audio or video so that software can later tell the content was made by AI — like a banknote's watermark, but undetectable to the eye. techniques and detectors developed by Google DeepMind for content generated by Google's models. For images and video it embeds an imperceptible signal designed to survive common edits such as cropping, filtering and compression; for audio it does the same in the waveform; for text it uses 'tournament sampling', which nudges token choices according to a secret key without measurably reducing quality, though detection weakens on short, factual or heavily rewritten passages. Detection reports a likelihood rather than a certainty. As of May 2026 Google said it had watermarked over 100 billion images and videos, offered verification of images, video and audio in its Gemini app and a detection API to partners, and other firms, including OpenAI and NVIDIA, had announced adoption alongside Content CredentialsContent Credentials, prop. n.: A tamper-evident 'nutrition label' attached to a photo, video or document recording where it came from — which camera or AI tool made it and how it was edited — that anyone can inspect. metadata.
The newsroom ran the viral photo through a SynthID check before publishing it.
Origin
Announced by Google DeepMind in August 2023 for images from its Imagen model, then extended to audio and, in 2024, to text and video. Sumanth Dathathri and colleagues described the text method in Nature (October 2024), when it was also released as open source; a public SynthID Detector portal followed in 2025.
Related
- AI watermarkingAI watermarking, n.: Hiding an invisible signal in AI-generated text, images, audio or video so that software can later tell the content was made by AI — like a banknote's watermark, but undetectable to the eye.
- Content CredentialsContent Credentials, prop. n.: A tamper-evident 'nutrition label' attached to a photo, video or document recording where it came from — which camera or AI tool made it and how it was edited — that anyone can inspect.
- C2PAC2PA, abbr.: An industry group, and the technical standard it publishes, for attaching tamper-evident records of origin and editing history to digital photos, video, audio and documents.
- deepfake detectiondeepfake detection, n.: Using software, often itself AI, to tell whether a video, photo or voice recording has been faked or generated by AI rather than captured from reality.
- Google DeepMindGoogle DeepMind, prop. n.: Google's main artificial intelligence laboratory, formed in 2023 by joining DeepMind and Google Brain. It builds the Gemini models and scientific AI systems such as AlphaFold.
Sources
- Dathathri, S. et al. (2024). Scalable watermarking for identifying large language model outputs. Nature 634, 818–823.
- InfoQ (2026). Google Expands SynthID Adoption for AI Watermarking, Previews Content Detection API. May 2026. infoq.com/news/2026/05/google-synthid-content-detection
Mentioned in
AI watermarkingAI watermarking, n.: Hiding an invisible signal in AI-generated text, images, audio or video so that software can later tell the content was made by AI — like a banknote's watermark, but undetectable to the eye., AI detectorAI detector, n.: Software that scores how likely it is that an AI, not a person, produced some text or media; for writing, its verdicts are wrong often enough that many schools no longer trust them., AI-generated content labelingAI-generated content labeling, n.: Marking text, images, audio or video as made by AI — with a visible notice, hidden data in the file, or both — so that people and platforms can tell synthetic content from real., Content CredentialsContent Credentials, prop. n.: A tamper-evident 'nutrition label' attached to a photo, video or document recording where it came from — which camera or AI tool made it and how it was edited — that anyone can inspect., Nano BananaNano Banana, prop. n.: The nickname, later Google's official branding, for its Gemini image models, a viral hit in 2025 for editing photos from simple instructions while keeping people looking like themselves., synthetic mediasynthetic media, n.: Images, video, audio or text created or substantially altered by AI rather than recorded or written directly by people, whether for art, entertainment, business or deception., VeoVeo, prop. n.: Google's video-generating AI, which since 2025 creates short clips complete with matching sound and dialogue from a written prompt., C2PAC2PA, abbr.: An industry group, and the technical standard it publishes, for attaching tamper-evident records of origin and editing history to digital photos, video, audio and documents.