SynthID Detector
SynthID is four systems under one name: text, image, audio and video. Only some of them can be checked by anyone outside Google, and none of the checks are a browser paste.
- Watermark type
- Statistical text watermark
- Hidden characters
- No
- Official detector
- Partial: Partial Detector portal is waitlist-only
- Public API
- Partial: Partial Waitlist or restricted access
- DetectMark support
- DetectMark support: Unavailable No public detector to call
- Removal
- Unavailable: Unavailable Rewrite only; not a character strip
- Watermark type
- Pixel watermark + C2PA metadata
- Hidden characters
- No
- Official detector
- Partial: Partial Checks in the Gemini app and Google Search
- Public API
- Partial: Partial Waitlist or restricted access
- DetectMark support
- DetectMark support: Available Reads C2PA locally
- Removal
- Partial: Partial Metadata strips; pixels stay
What SynthID actually is
For text, it is the same family of technique as Claude's mark: a keyed bias in token selection, introduced in a 2024 Nature paper from Google DeepMind. For images it is different in kind. SynthID Image lives in the pixel values, which is why it survives a screenshot, a JPEG re-encode, a resize and a crop.
The open-source SynthID-Text repository publishes the code but not Google's production key. It can score text you watermarked yourself. It cannot score Gemini output.
Where verification exists
Image verification is live for people using Google Search surfaces, is rolling out in Chrome, and is in limited preview as a Cloud API. Text verification runs through the SynthID Detector portal, which is waitlist-only for journalists, media professionals and researchers.
One provider does expose an open check: OpenAI's Verify tool and Content Provenance API verify the SynthID that OpenAI itself embeds in its images and audio. That covers OpenAI-originated SynthID only; it cannot check Gemini output, and Google's own portal is still waitlist-only. None of this is a public paste-and-check page for SynthID in general, which is why SynthID rows in DetectMark report Partial or Unsupported rather than a verdict. The one live check on this site is the OpenAI one: the OpenAI image verifier sends the original file to OpenAI's verifier after an explicit consent click, and the Image tab below does the same after a local scan. That check covers OpenAI-supported signals only: a not-detected result says nothing about other providers or about a file whose marks were stripped.
What this page checks locally
Drop an image and DetectMark reads the file: Content Credentials, XMP, EXIF, GPS and PNG text chunks. C2PA often ships alongside SynthID, so its presence is a useful signal about where a file came from. It is not the same signal, and the result panel keeps them apart.
FAQ
Not directly. Google's SynthID Detector portal is waitlist-only for journalists, media professionals and researchers. There is no public paste-and-check page for SynthID Text.
Partially. The image tool reads C2PA Content Credentials that often ship alongside SynthID. For Gemini images, pixel verification is available in Google Search surfaces and rolling out in Chrome. For OpenAI images, OpenAI's Content Provenance API verifies the SynthID OpenAI embeds, and only that; DetectMark can send the original file to it after you consent.
Google says essentially all Gemini text is watermarked. An independent test on August 23, 2026 found the signal in Gemini 3.5 Flash output.
Same technique family, different keys. Google's detector cannot read Anthropic's key, and the reverse is equally true.
Only for text you watermarked with your own key. It cannot read Google's production Gemini key.
Every check runs in your browser: how it works. The limits are written down too: what DetectMark cannot remove.