SynthID Explained

Short answer

SynthID is Google's invisible watermark for AI content. It is built into images, video, audio and text made with Google's AI models, and only SynthID's own detectors can read it. You check for it in the Gemini app or on Google's SynthID Detector portal. A positive result is strong evidence. A negative one only rules out Google.

Most watermarks are something you see. SynthID is something you do not. It sits inside the picture and the sound, and it is meant to stay there after the clip is cropped, filtered and squashed by a dozen re-uploads. That makes it one of the more useful signals for one specific question: did Google's AI make this? It is also why people misread it, because that is the only question it answers.

What SynthID is

Google DeepMind describes SynthID as a tool that embeds digital watermarks directly into AI-generated images, audio, text or video. The watermarks are imperceptible to people but detectable by SynthID's own technology. For images and video, Google says the mark is added the moment content is created, does not change quality, and is designed to stand up to modifications like cropping, adding filters, changing frame rates, or lossy compression.

Google's May 2025 announcement of the detector portal says SynthID covers content from its Gemini, Imagen, Lyria and Veo models, and that over 10 billion pieces of content had been watermarked by that date. Veo is Google's video model, so if you suspect a clip came from Veo, SynthID is the first thing to check. The free Veo video checker covers the visual side of that question.

The Gemini help page draws the line between the two invisible signals neatly: watermarks are embedded directly into content, while metadata adds context to original files. One is woven into the pixels. The other rides along in the file and can fall off.

Where to check for it

Google gives you two ways to check.

The Gemini app. Google's help page explains that signed-in users can upload an image, video or audio file and ask whether it was created or edited by Google AI. As of November 2026 that page lists these limits for video: one file at a time, 100 MB or less, under 90 seconds long, and a rolling quota of about 10 video checks (up to 5 minutes total) per 24 hours. For video, Gemini says which parts of the clip carried the watermark. DeepMind's SynthID page adds that Gemini, Search and Chrome will check for a SynthID watermark when you upload a file and ask.

The SynthID Detector portal. Google launched the portal in May 2025 for early testers, then opened it up. DeepMind's SynthID page now says that anyone can check if an image, video, or audio file was made with AI from Google or partners, including OpenAI, NVIDIA, Kakao, and soon, Apple, and links to the portal at synthid.com. The portal highlights the parts of a file most likely to carry the watermark.

How to run a useful check:

1. Use the best copy you can get: the original download beats a screen recording.
2. Trim to the part you care about if the clip is long.
3. Ask a direct question: "Was this video created or edited by Google AI?"
4. Read which segments were flagged, not just the headline answer.

What it cannot tell you

Read the result as narrowly as Google writes it. The help page puts it this way:

• If a watermark is found, all or part of the image or video was created or edited by Google's AI models.
• If none is found, the file was not made or edited by Google AI, but it could have been made by other AI systems.
• Some results are unclear, for example very simple content, or an edit too small to carry the mark.

Four limits follow from that.

It only knows its own marks. The Gemini help page says Gemini can currently only recognize content created by Google AI tools. A clip from a tool that does not use SynthID will always come back clean.

Partial means partial. A flagged segment shows Google AI touched that part. It does not say the whole clip is synthetic, or that the rest is real.

Heavy handling can wear it down. Google says the watermark usually survives resizing, recolouring and compression, but after many alterations it may not be detected.

It says nothing about truth. A real event can be edited with AI. A fully synthetic clip can be labelled honestly as fiction. SynthID answers "which tool touched this", not "is the claim in this clip true".

This site's checker cannot read SynthID. It works differently, by scoring what the frames look like, which is why the two make a good pair.

Clips with no watermark: other ways to check

Most suspicious clips will come back with no SynthID mark, either because another tool made them or because they are real. Here is what to do next.

Run a file check. The free detector on the homepage reads the file in your browser for Content Credentials under the C2PA specification and for AI labels that use the IPTC digital source type codes, and shows still frames from the clip for you to inspect. It does not read SynthID. You get one of four verdicts: AI generated, Likely AI, Likely real, or No AI signs found. It is free with no signup. "No AI signs found" is one signal among several, not proof the clip is real. If you want a second opinion, the detector comparison lines up free tools on what they show you.

For a frame-by-frame AI score, DeepfakeDetector.ai has a free account with 50 scans a month for clips up to 2 minutes, and it also checks images and voice.

Look for content credentials. The Gemini app also reads C2PA Content Credentials, and the Content Credentials Verify tool reads them too. Credentials can show who signed a file and whether AI tools were used. The Sora watermark explainer covers how to read them and why they often go missing.

Check the visuals and the source. Hands, text, physics and shadows still give generated video away, and a reverse image search on a clear frame often finds the first upload. Google's own help page recommends both: look for visual artifacts and use reverse image search.

For clips you suspect came from Google's own video tools, the Veo video checker pairs a free file check with the SynthID steps above. Guide maintained by Tessa Ostrowski.

Sources

1. Google DeepMind: SynthID
2. Google, 20 May 2025: SynthID Detector, a new portal to help identify AI-generated content
3. Google, Gemini Apps Help: Verify AI-generated images, videos, and audio
4. Google: SynthID Detector portal
5. Content Credentials: Verify tool
6. C2PA: C2PA Technical Specification 2.1
7. IPTC: Digital Source Type NewsCodes

All accessed 9 November 2026.