SynthID Detector: How to Check for Google's AI Watermark
Google's SynthID Detector is now free for everyone. It is excellent at one thing, spotting Google's invisible watermark, and blind to almost everything else. Here is how to use it, how to read the result, and what to check when it finds nothing.
SynthID Detector is Google's free tool at synthid.com for checking whether an image, video or audio file carries a SynthID watermark, the invisible signal Google and partners such as OpenAI, NVIDIA, Kakao and Apple embed in AI content. Upload the original file, or ask the Gemini app "Is this AI-generated?". A detected watermark is strong evidence of AI. No watermark proves nothing, so follow up with a general AI detector and a C2PA check.
- Free for everyone since October 7, 2026 at synthid.com, plus built-in checks in Gemini, Search and Chrome.
- Covers images, video and audio from Google AI and named partners only.
- A detected watermark is strong evidence. No watermark is not evidence the file is real.
- Pair it with two other checks: an AI image detector for any generator, and a C2PA check for Content Credentials.
- DeepfakeDetector.ai does not read SynthID. It covers the gaps SynthID leaves.
- What is SynthID?
- What is SynthID Detector?
- How to check an image or video for SynthID
- What a SynthID result does and does not mean
- What SynthID can't detect
- SynthID vs C2PA vs an AI detector
- What to check when SynthID finds nothing
- Does SynthID work on text?
- FAQ
- Conclusion: a great yes, a weak no
What is SynthID?
SynthID is Google DeepMind's invisible watermark for AI-generated content. When one of Google's models creates an image, a video clip or audio, SynthID weaves a signal into the content itself, into the pixels or the sound, that people cannot see or hear but a detector can read. Google says the watermark is built to stay detectable after common changes such as compression and cropping.
Google launched SynthID for images in 2023 and has since extended it to text, audio and video, including content from its Gemini, Imagen, Lyria and Veo models (Google, May 2025). By October 2026 Google said it had watermarked more than 180 billion images and videos and 240,000 years of audio (Google, October 2026).
The key point for anyone checking a file: SynthID is not a general AI detector. It answers one narrow question, whether this content carries a SynthID watermark. That is very reliable evidence when the answer is yes, and almost no evidence when the answer is no.
What is SynthID Detector?
SynthID Detector is Google's free web portal for checking files for the watermark. Google first announced it in May 2025 as a waitlist tool for journalists and researchers. On October 7, 2026, Google opened it to everyone, worldwide, in English, at synthid.com (Google).
| Question | Answer (as of November 2026) |
|---|---|
| Where | synthid.com, plus built-in checks in the Gemini app, Google Search and Chrome |
| What it checks | Images, video and audio files |
| Whose content | Google AI, plus partners Google names as OpenAI, NVIDIA, Kakao and Apple |
| Cost | Free |
| Language | English interface at launch |
| What it cannot tell you | Whether content from any other generator is AI |
Google says its built-in verification features in Search, the Gemini app and Chrome handle more than one million requests a day (Google). The portal adds a place to upload a file directly, and it highlights which parts of an image or which seconds of audio carry the watermark.
How to check an image or video for SynthID
- Get the file itself. Download the image, video or audio. A screen recording or a photo of a screen is a new file and may not keep the watermark as well as the original.
- Upload it to SynthID Detector. Go to synthid.com and upload the file. The result says whether a SynthID watermark was found and, if so, where.
- Or ask the Gemini app. Upload the image or clip and ask "Is this AI-generated?" or "Was this created with Google AI?". Image checks launched on November 20, 2025 (Google) and video checks on December 18, 2025, for files up to 100 MB and 90 seconds (Google). For video, Gemini reports the audio and the visuals separately, for example "SynthID detected within the audio between 10-20 secs".
- Then check what SynthID cannot see. If no watermark is found, run the file through a general AI image detector and a C2PA checker before you decide anything.
What a SynthID result does and does not mean
| Result | What it means | What to do |
|---|---|---|
| Watermark detected | Strong evidence the content, or the highlighted part of it, was made or edited with Google AI or a participating partner | Treat it as AI-generated. Note which parts were flagged. |
| No watermark detected | The file has no SynthID signal. It may be real, or made by any generator outside the program | Not proof of anything. Run a pixel-based AI check and look for C2PA credentials. |
| Part of the file flagged | Some segments or regions carry the watermark, others do not | Common for edited media, such as AI audio over real video. Judge the flagged part on its own. |
The "no watermark" row is the one people misread. Plenty of AI images and videos online come from models that never embed SynthID, including open-source models people run on their own computers. A clean SynthID result on one of those files says nothing about whether it is real.
What SynthID can't detect
- Content from non-participating generators. SynthID only exists in content from Google and the partners that embed it. Everything else is invisible to it.
- Content made before a partner joined. A partner's older output will not carry the watermark.
- Real content that was faked by other means. A face swap built with an open-source tool, a cheap edit, or a real photo with a false caption all pass with no signal.
- Published error rates. We could not find public false positive or false negative rates for the portal as of November 2026, so treat a result as strong evidence rather than a guarantee.
SynthID vs C2PA vs an AI detector
There are three different ways to ask "is this AI?", and they fail in different places. That is why serious verification uses more than one.
| SynthID | C2PA Content Credentials | AI detector (pixel analysis) | |
|---|---|---|---|
| What it is | Invisible watermark inside the pixels or audio | Signed provenance record in the file's metadata | A trained model that reads the content itself |
| Who adds it | Google and partners, at creation | Participating apps, cameras and generators | Nobody; it works on any file |
| Survives screenshots and uploads | Designed to survive common edits | Usually stripped | Works, with lower confidence on degraded files |
| Covers other generators | No | Only those that sign | Yes |
| Answer you get | Watermark found or not | Tool named, with edits | A probability and a verdict |
| Where to check | synthid.com, Gemini app | C2PA checker, contentcredentials.org | AI image detector |
What to check when SynthID finds nothing
To be clear about our own tools: DeepfakeDetector.ai does not read SynthID watermarks. Only Google's tools can. What we offer are the two checks that cover SynthID's blind spots.
- The free AI image detector scores the pixels of any image for AI generation and editing, whichever model made it, and returns a verdict and an AI score. No signup for up to 10 checks a day.
- The free C2PA checker reads Content Credentials and metadata and names the generator or app when the file still records it, for example OpenAI's image models or Adobe Firefly.
- For video and voice, a free account covers the AI video detector and the AI voice detector, 50 scans a month.
A practical order: SynthID first if you suspect a Google or partner model, then the C2PA check, then the AI score. If all three come back empty on a file that matters, look for the original source and treat the content as unverified rather than real.
Does SynthID work on text?
Google built a text version of SynthID, which adjusts word choices during generation in a way a detector can recognise. Google open-sourced SynthID Text, and it shipped in Hugging Face Transformers version 4.46.0 on October 23, 2024, so developers can watermark their own models' output (Hugging Face). Text watermarks weaken after heavy rewriting or translation. The public SynthID Detector launched in October 2026 is described as a checker for images, video and audio files.
FAQ
What is SynthID Detector?
SynthID Detector is Google's free web portal at synthid.com for checking whether an image, video or audio file carries a SynthID watermark. Google opened it to everyone on October 7, 2026. It detects content made with Google AI and with partners Google names as OpenAI, NVIDIA, Kakao and Apple.
How do I check if an image has a SynthID watermark?
Upload the original file to synthid.com, or upload it to the Gemini app and ask "Is this AI-generated?". Both check for SynthID. If no watermark is found, run a general AI image detector and a C2PA check, because SynthID only covers participating generators.
Does no SynthID watermark mean the image is real?
No. SynthID only appears in content from Google and participating partners. Images from other generators, including open-source models, never carry it, so a clean result is not evidence that a file is real.
Is SynthID Detector free?
Yes. Google offers SynthID Detector free at synthid.com, and SynthID checks are also built into the Gemini app, Google Search and Chrome.
What is the difference between SynthID and C2PA?
SynthID is an invisible watermark embedded in the pixels or audio by Google and its partners. C2PA is signed provenance metadata stored in the file, which anyone can read but which uploads and screenshots usually strip. They answer different questions and work best together.
Can DeepfakeDetector.ai detect SynthID?
No. Only Google's tools read SynthID watermarks. DeepfakeDetector.ai offers the checks that cover SynthID's blind spots: a free AI image detector that scores any image, and a free C2PA checker that reads Content Credentials and metadata.
Conclusion: a great yes, a weak no
SynthID Detector is the best way to confirm that a file came from Google's AI or one of its partners, and it is now free for anyone at synthid.com. Its blind spot is everything else. When it finds a watermark, believe it. When it finds nothing, keep checking.
Cover the gap in one minute. Run the free AI image detector and the C2PA checker on the same file. For more on how these signals work, read how AI image detectors work.