SynthID is Google DeepMind's invisible watermark for AI-generated content. It marks images, video, audio and text made with Google AI, and Google's public SynthID Detector, open to everyone since 2026-10-07, reads it for images, video and audio only.
How it works by media type
- Images, video and audio. The signal is embedded in the content. Google's detector and the Gemini app can check files, but only for content made with Google AI or its named partners.
- Text. Google says SynthID watermarks text from the Gemini app and web. Detection is probabilistic, and the signal fades with heavy rewriting or translation. There is no public text verifier, so see the watermark status.
What a missing mark does and does not mean
No watermark found means the checker did not find one. It does not show a human made the content, because other AI tools can produce images and text without SynthID. It is also not proof either way for edited files.
Checking images
Pictures from Google tools are the most common SynthID question. For an image, use an image detector that does not depend on a watermark, and check the file's embedded metadata. The Gemini detector facts sheet lists what Google's own tools do and do not cover.
Checking text
For text, use GeminiDetect to see an AI verdict and Gemini style similarity. It does not read SynthID. Its engine, Veread, comes from Fileverity, and the best Gemini AI detectors page compares it with other tools.
Sources: Google DeepMind SynthID pages; Google blog, 2026-10-07; Gemini Apps Help.