What this checker does, and what it can't
SynthID is an invisible watermark that Google DeepMind builds into content made with Google's AI models (Gemini, Imagen, Veo and Lyria), and that OpenAI has used on its own images since May 2026. It lives in the pixels or the sound itself. Only the companies that add it can detect it (Google in Gemini and Search, OpenAI in its Verify tool), so no third-party website, including this one, can read it for you.
What a browser can read is the metadata that travels with the file, and Google writes quite a lot of it. This checker reads it on your device and tells you which markers are there, where in the file they sit and what they mean. Then it shows you how to check the watermark itself in Gemini.
How to use it
- Drop an image, video or audio file on the box, choose one, or paste a copied image. Use the original download if you can: screenshots and files saved from social networks usually lost their metadata.
- Read the result. Google AI markers found means the file says it was made with Google AI; Google AI editing markers found means it says Google AI edited it; AI markers found, none from Google means another company's labels. No AI markers in the metadata means only that: nothing in the metadata, not “not AI”.
- Follow the Gemini steps under the result to check the SynthID watermark. Use Share result to send someone the list of markers (the link carries no image data and no file name), or Copy text report for a fact-check note.
The markers it looks for
- C2PA Content Credentials: a signed record of who made the file and how. Google has added them to Nano Banana Pro images since about November 2025, and signs images and Veo videos made with its models on Google Cloud as Google LLC, with “Google Media Processing Services” as the app. Pixel 10 photos are signed by the camera, and Google Photos adds credentials when you edit with AI. The checker verifies the signature, and says “signed by Google” only when the signer's certificate is on the C2PA trust list.
- IPTC digital source type:
trainedAlgorithmicMedia(“created using generative AI”) orcompositeWithTrainedAlgorithmicMedia(“edited using generative AI”). Google uses both. - Credit line: Gemini images carry the IPTC credit “Made with Google AI”.
- Software and AI-system fields that name Gemini, Imagen, Nano Banana, Veo, Lyria or another company's generator, and the newer IPTC “AI system used” field.
- Other companies' markers for context: C2PA from OpenAI (ChatGPT, Sora), Adobe Firefly and Microsoft, and IPTC labels from Midjourney, Meta and others.
- Pixel camera data, which says nothing about AI on its own but explains why an edited photo still has camera EXIF.
Plain metadata (IPTC, credit lines, software fields) is unsigned: anyone can add or remove it. Only C2PA is signed, and even a valid signature proves who signed the file, not that its contents are true. For the full signed history, open the file in the Content Credentials checker; for evidence from other generators, such as Stable Diffusion prompts, use the AI image detector.
Metadata versus the watermark
Metadata and SynthID fail in opposite ways. Metadata is easy to read and easy to lose: a screenshot, a WhatsApp forward or an EXIF remover deletes it, and none of that touches SynthID, which Google says is designed to stand up to cropping, filters, changed frame rates and lossy compression. So stripped metadata doesn't remove the watermark, and a file with no markers here can still be watermarked. The watermark is hard to lose but hard to check: you need Gemini, a Google account, and for Gemini to recognise the model. It only recognises Google's own tools, not the SynthID that OpenAI or other partners add.
How to check SynthID in Gemini
- Open gemini.google.com or the Gemini app for Android or iPhone, signed in with a personal Google account (or a work account where Gemini is enabled).
- Add the file: one image, video or audio file up to 100 MB. Videos must be under 90 seconds and audio under 1 hour.
- Ask “Was this created or edited by Google AI?” or “Is this AI-generated?”. Gemini tells you whether it found SynthID, and at gemini.google.com and in the Android app it also reads the Content Credentials.
Google allows about 10 image, 10 video (5 minutes in total) and 10 audio checks (3 hours in total) in any 24 hours. For a screenshot, crop tight around the picture first. Answers like “likely too small an edit” mean the result is inconclusive. Since 19 May 2026 Google Search can also check SynthID (it is rolling out to Chrome), and Google Cloud offers an AI Content Detection API to trusted partners. The SynthID Detector portal, which highlights the parts most likely to be watermarked, is open only to a waitlist of journalists, media professionals and researchers. For images and audio from ChatGPT or OpenAI's API, use OpenAI's Verify tool, a public preview that checks C2PA and SynthID in images, and SynthID in audio.
Troubleshooting
- A Gemini image shows no markers: it was probably copied, screenshotted or re-saved. Download the original from Gemini again, or check the watermark in Gemini.
- “Valid, signer not on a trust list”: the credentials are intact, but their signer isn't verified, so a name like Google in them is only a claim.
- Invalid credentials: the file was changed after signing, usually by an app that doesn't support C2PA. The markers describe an earlier version.
- A long video: this checker reads long files too (Content Credentials up to 1 GB), but Gemini only checks clips under 90 seconds and 100 MB, so trim the part you care about first.
Sources
- Gemini Apps Help: Verify AI-generated images, videos, and audio, read 26 September 2026
- Google: image verification in the Gemini app, C2PA in Nano Banana Pro images, November 2025
- Google: video verification in the Gemini app (up to 100 MB and 90 seconds), 18 December 2025
- Google: SynthID checks in Search (rolling out to Chrome), C2PA checks in Gemini, AI Content Detection API, 19 May 2026
- Google: SynthID Detector portal and waitlist, 20 May 2025
- Google DeepMind: SynthID, read 26 September 2026
- Google Cloud: Content Credentials on generated images and videos, updated 25 September 2026
- Google: C2PA in the Pixel 10 camera and Google Photos, 10 September 2025
- IPTC: Google AI transparency based on IPTC standards, 28 October 2024
- IPTC NewsCodes: Digital Source Type vocabulary, modified 26 September 2024
- OpenAI: SynthID in ChatGPT, Codex and API images (audio added 31 July 2026), 19 May 2026
- OpenAI: Verify OpenAI-generated content (public preview), read 26 September 2026