Google SynthID Detector Explained: How to Check AI Images, Video and Audio
Quick answer
Google’s SynthID Detector is a verification tool that checks an image, video or audio file for an invisible SynthID watermark. A positive result indicates that Google’s system found a watermark associated with Google or a supported partner’s AI-generated or AI-edited content. A negative result does not prove that a file was made by a human or that it contains no AI content.
AI-generated media is becoming difficult to judge by appearance alone. A realistic image may be created by an AI model, while an authentic photograph may have been heavily edited. Google’s response is SynthID, a watermarking system designed to add an imperceptible signal to selected AI-generated content.
On October 7, 2026, Google DeepMind said it was expanding the SynthID Detector to everyone globally in English. The company says the detector can check images, video and audio created by Google or participating partners, including OpenAI, NVIDIA and Kakao, with Apple support planned. [web:760]
But SynthID is not a universal AI-content detector. It looks for a specific watermark. That distinction is the most important thing to understand before using the result for journalism, schoolwork, moderation or fact-checking.
What is Google SynthID Detector?
Google SynthID Detector is a verification portal that checks digital media for an invisible SynthID watermark. The watermark is designed to be difficult for people to notice and resilient to common changes such as compression, cropping, filters or changes to video frame rates.
Google says SynthID has been used across images, video and audio, and that more than 180 billion images and videos plus 240,000 years of audio had been watermarked through its systems by the October 2026 announcement. These are Google’s reported figures, not an independent measurement. [web:760]
The detector answers a narrow question: does this file contain a recognizable SynthID signal? It does not automatically determine who created the file, whether the entire file is AI-generated or whether a file without a watermark is definitely human-made.
How SynthID watermarking works
SynthID embeds a digital watermark into generated content during creation. In images and video, the signal is placed into the visual content without creating a visible logo. In audio, Google describes the watermark as inaudible to listeners.
The system is designed to remain detectable after common transformations. Google says image and video watermarks are designed to withstand cropping, filters, frame-rate changes and lossy compression. Audio watermarks are designed to withstand changes such as noise, MP3 compression and speed adjustments. [web:748]
Durability does not mean perfection. Google’s own help documentation says that repeated or extensive alterations can make detection harder, and some content may not contain enough detail for reliable watermarking. [web:749]
| Media type | Watermark behavior | What can affect detection |
|---|---|---|
| Image | Invisible signal embedded during supported AI generation or editing. | Very small edits, heavy transformations or unsupported generation tools. |
| Video | Watermark may be detected in particular portions of a file. | Editing, re-encoding, short clips or unsupported content. |
| Audio | Inaudible signal embedded in supported AI-generated audio. | Noise, repeated processing or unsupported audio generation. |
How to check a file with SynthID
Google’s official announcement refers to the public SynthID Detector portal. Google also documents SynthID verification inside Gemini for signed-in users. Because interfaces can change during rollout, use the official Google page rather than a third-party clone.
General verification workflow
- Open Google’s official SynthID Detector or Gemini verification experience.
- Sign in if the service asks for an account.
- Upload one image, video or audio file.
- Wait for the verification result.
- Read the result as a watermark signal—not as a universal AI verdict.
- Keep the original file and record the date, file type and result if the check matters.
For Gemini’s documented verification workflow, Google says users can upload a file and ask a question such as “Was this image, video or audio created or edited by Google AI?” Users can also enter @synthid to start the verification process. [web:749]
File limits in Gemini verification
Google’s Gemini Help page lists specific limits for its verification workflow. These limits should not automatically be assumed to apply to every version of the public detector portal.
| File type | Documented Gemini limit |
|---|---|
| Image | One file at a time; file size should be 100 MB or less. |
| Video | One file at a time; each video should be shorter than 90 seconds. |
| Audio | One file at a time; each audio file should be shorter than one hour. |
| Daily quota | Google Help lists approximately 10 image checks, 10 video checks totaling 5 minutes, and 3 hours of audio. |
These limits come from Google’s Gemini verification documentation and may change. Check the current interface before designing a large research or moderation workflow. [web:749]
How to interpret the result
The result needs careful wording. A positive watermark result and a negative result do not carry the same meaning.
| Result | What it supports | What it does not prove |
|---|---|---|
| SynthID detected | Google found a watermark associated with supported AI-generated or AI-edited content. | It does not prove that every part of the file was generated by AI or identify the exact human operator. |
| SynthID not detected | No recognizable SynthID signal was found in the checked file. | It does not prove the file is human-created or free from other AI systems. |
| Not enough details | The content may not contain enough information for a reliable watermark check. | It is not proof of either human or AI creation. |
| Unclear | The available signal is insufficient for a confident classification. | It should not be presented as a final authenticity verdict. |
Can SynthID detect every AI image?
No. SynthID is not designed to detect every AI-generated file from every model. It checks for a watermark associated with supported systems and partners.
A file generated by an unsupported tool may return no watermark even if it was entirely AI-generated. A screenshot or heavily edited copy may also produce an uncertain result. Google’s own documentation says a missing watermark can mean the content was not created or edited by Google AI, but it could still have been created by another AI system. [web:749]
SynthID versus ordinary AI detectors
SynthID and statistical AI detectors use different approaches. SynthID checks for a designed watermark, while ordinary AI detectors typically estimate whether content resembles AI-generated output.
| Approach | How it works | Main limitation |
|---|---|---|
| SynthID verification | Looks for a specific embedded watermark. | Works only when a recognizable supported watermark exists. |
| Visual inspection | Looks for strange text, hands, shadows, lighting or repeated patterns. | Humans and AI models can both miss subtle errors. |
| Metadata review | Checks file information, camera data or editing traces. | Metadata can be removed or changed. |
| Reverse image search | Looks for earlier versions or the original source. | It may not find newly created or private content. |
For important verification, combine the watermark result with source research, reverse search, metadata review and direct confirmation from the creator. No single method should be treated as perfect.
Our editorial testing plan
We are not claiming a measured accuracy percentage from a small sample. A responsible hands-on test would need clearly labelled files, the same upload process and a record of every transformation.
Repeatable test setup
- One Google-generated image.
- One image from a supported partner, where permission and access are available.
- One normal camera photograph.
- One cropped or compressed version of an image.
- One short supported AI-generated video.
- One human-recorded audio file and one supported AI-generated audio file.
For each file, record the source, creation tool, file type, file size, edits, upload date and result. A six-file experiment can show workflow behavior, but it cannot establish universal detection accuracy.
Privacy and responsible use
Uploading media to an online verification service can expose sensitive information. Do not upload private documents, confidential business recordings, unreleased product material or personal audio without checking the service’s terms and your organization’s policy.
- Use a copy when possible, not the only original.
- Remove unrelated personal information before upload.
- Do not upload someone else’s private recording without permission.
- Keep the original file and verification result together.
- Do not accuse a person of deception based only on a negative or unclear result.
- For legal, academic or newsroom decisions, use multiple sources of evidence.
TechZila analysis: watermarking is context, not a verdict
Google’s move matters because AI media is becoming a provenance problem, not just a quality problem. People increasingly need to know where a file came from, which system created it and whether the content was edited later.
SynthID addresses one part of that problem: identifying a signal created by participating AI systems. It does not solve the wider question of authenticity. A real photograph can be misleading, an AI image can be used honestly, and a file without a watermark can still be synthetic.
Our view is that SynthID will be most useful as a first verification layer for journalists, educators, creators and platforms. Its responsible use depends on careful wording: “a watermark was detected” is stronger and more accurate than “this proves the entire file is AI-generated.”
TechZila verdict
Google SynthID Detector is a useful provenance tool for supported AI media, but it is not a universal AI lie detector. Use it to find evidence of a supported watermark, then combine the result with source checking and context.
Frequently asked questions
What is Google SynthID Detector?
It is a tool that checks images, video and audio for an invisible SynthID watermark associated with supported AI-generated or AI-edited content.
Does SynthID detect all AI-generated content?
No. It is designed to detect supported watermarks. A file made with an unsupported AI system may not contain a detectable SynthID signal.
What does “SynthID detected” mean?
It means the system found a watermark associated with Google or a supported partner’s AI content. It does not necessarily mean every part of the file was AI-generated.
What does “SynthID not detected” mean?
It means no recognizable SynthID watermark was found. It does not prove that the file was made entirely by a human.
Can SynthID check audio?
Yes. Google describes SynthID watermarking for supported AI-generated audio and documents audio verification in Gemini. File limits and supported formats can vary.
Can I check a video?
Yes. Google says SynthID can check video, and Gemini may identify portions where a watermark was detected. The result may not mean every frame contains the watermark.
Is SynthID free?
Google has expanded the detector to everyone globally in English, but sign-in requirements, quotas and interface availability may vary. Check the official detector page before relying on it for regular use.
Sources and update policy
This article uses Google DeepMind’s announcement, Google’s SynthID overview and Google’s Gemini Help documentation. We will update the article when Google publishes changes to supported partners, file limits, quotas or public access.
Last checked: October 8, 2026.
0 Comments