Quick answer: Google's SynthID Detector became available to everyone in English globally on October 7, 2026, according to Google DeepMind. It is designed to help identify supported media carrying SynthID watermarks. A positive result can provide useful provenance evidence, but a negative result does not prove that an image, video, or audio recording is authentic.
A realistic AI-generated video can look convincing enough to be shared before anyone checks where it came from. The same problem affects photographs, voice recordings, music, and edited clips. As generative AI becomes easier to use, media verification needs to rely on more than visual intuition.
One approach is to check for a technical provenance signal. Google DeepMind's SynthID system embeds imperceptible watermarks in supported AI-generated media. On October 7, 2026, Google announced that its SynthID Detector was expanding from earlier professional access to public availability in English globally.
The tool can help people investigate whether supported images, videos, or audio files were created or edited using Google's AI technology or participating partners. Its scope is important: it is not a universal detector for every AI system, and it should not be treated as a final verdict on whether an entire piece of media is genuine.
What Is SynthID?
SynthID is a digital watermarking technology developed by Google DeepMind. Rather than relying only on a visible label, the system embeds a signal designed to remain detectable within supported content without making an obvious change to the media's appearance or sound.
For images, a watermark may be embedded into the generated visual content. For video, verification can involve examining visual content and, where supported, audio. For audio, the watermark is intended to be detectable even though listeners cannot ordinarily hear it.
The purpose is not to prevent people from creating synthetic media. It is to make it easier to obtain information about the origin of content when a compatible watermark is present.
Google has described SynthID as part of a broader effort to improve transparency around AI-generated content. In its October 7 announcement, the company said it had watermarked more than 180 billion images and videos and an amount of audio equivalent to 240,000 years. Those are figures reported by Google, not an independent measurement of every AI-generated item circulating online.
What Changed in October 2026?
Google announced that the SynthID Detector was expanding to public use in English globally. The company said the service could help people check whether an image, video, or audio file was made with Google AI or supported partner technology.
Google identified OpenAI, NVIDIA, and Kakao among the partners covered by its announcement, with Apple described as a future addition. That statement should not be interpreted as meaning that every output from those organizations carries SynthID or that the detector can identify every kind of AI-generated content.
Google also described existing verification features in Search, the Gemini app, and Chrome. These related tools can provide additional ways to investigate supported content, but their precise capabilities and upload requirements may differ.
Important: A tool that checks for SynthID is checking for a particular provenance signal. It is not automatically checking every possible AI watermark, every metadata standard, or every manipulation technique.
How to Check a File
Step 1: Preserve the original file
If you are investigating a suspicious image, video, or audio recording, save the original file if it is available. Avoid repeatedly converting it, compressing it, or editing it before the first check.
Social networks and messaging services may resize images, recompress videos, strip metadata, or transform audio. Keeping the earliest available version helps preserve evidence that may be lost during later processing.
If the content was sent to you by another person, record where and when you received it. In a professional investigation, maintaining a documented chain of custody is more useful than keeping only a screenshot of the post.
Step 2: Use the official verification service
Open Google's official SynthID Detector through the Google DeepMind page listed in the sources below. Follow the current instructions to submit a supported file.
The service's supported file types, upload limits, and other restrictions should be checked on the live interface before you begin. Do not assume that the limits of one Google product automatically apply to another. For example, Google's earlier Gemini video-verification feature had its own upload constraints; those are not necessarily identical to the public detector's requirements.
Step 3: Read the result narrowly
If the service reports a SynthID watermark, read the result as evidence that the supported signal was detected. The result may help identify that some content was generated or edited with a supported AI system.
That does not necessarily establish who created the file, who uploaded it, whether the creator intended to mislead anyone, or whether every part of the media was synthetic.
A video may contain genuine footage combined with AI-generated audio or edited visual segments. A verification result should be interpreted in the context of the particular signal that was detected.
Step 4: Check the claim against the evidence
Suppose a social post claims that a video shows an actual event. If a supported watermark is detected, that may be important evidence about the clip's origin. It does not, by itself, explain the entire editing history or prove the caption is false in every respect.
Conversely, if no watermark is detected, do not conclude that the video must be authentic. The content could have been generated using another system, processed in a way that affects the signal, or created without a compatible watermark in the first place.
What a Positive Result Can Tell You
A positive result can help answer a limited but useful question: does the file contain a supported watermark associated with the relevant AI-generation or editing process?
That information can be valuable to journalists, researchers, educators, publishers, and people assessing viral media. It can also help creators document the use of supported AI tools in their own workflows.
However, the watermark does not establish the truth of the content's message. AI-generated media can illustrate a real event accurately, depict a fictional scene, or falsely imply that something happened. Provenance and factual accuracy are separate questions.
What a Negative Result Cannot Prove
A negative result means the service did not identify the supported signal it was checking for. It does not establish that the file was created by a camera, recorded in a particular place, or left unedited.
Several explanations may account for a negative result:
- The content was generated with a tool that does not use the watermark being checked.
- The file was edited or transformed, affecting the signal or metadata.
- The watermark was absent from the original output.
- The file format or particular media segment was not supported by the verification method.
- The signal was not detected in that particular check.
These possibilities mean that provenance tools should be used as part of a verification process, not as a replacement for it.
Combine Watermarks With Other Evidence
Professional verification often requires several independent checks. No single signal answers every question about the origin and accuracy of digital media.
Check the source of the post
Look for the earliest available publication. Is the account known to the organization or person it claims to represent? Is there an original source rather than a chain of reposts? Can the publisher explain where the file came from?
Inspect metadata carefully
File metadata may contain useful information about the software, device, or editing history. But metadata can be removed, changed, or missing for ordinary reasons. Its presence is not proof of authenticity, and its absence is not proof of manipulation.
Look for provenance standards
Some content uses provenance information associated with standards such as C2PA. These systems can record information about creation and editing in a verifiable structure. Their usefulness depends on implementation, the available credentials, and whether the information remains attached to the file.
Provenance metadata and invisible watermarking are related in purpose but are not the same mechanism. A file can carry one signal without carrying another.
Verify the event itself
For a newsworthy video, look for independent footage, statements from credible witnesses, reliable reporting, and records from relevant organizations. Reverse-image search and frame-level investigation may help identify an older clip that has been given a new caption.
Do not make a serious accusation based only on an automated detection result. A responsible report should describe the method used, the result obtained, and the limits of what that result establishes.
How Creators Can Use Provenance Tools
Creators can use verification technology as part of a transparent publishing workflow. If AI-generated imagery or video is used in a documentary, tutorial, advertisement, or educational presentation, keeping source files and creation notes can make later questions easier to answer.
Publishers should also maintain clear records of the distinction between real footage, AI-generated scenes, and substantial AI-assisted editing. That distinction matters even when the underlying creative work is lawful and non-deceptive.
For teams, a sensible process is to preserve originals, record the tools used, check supported provenance signals, and document any transformations made during editing. A final export should not be assumed to retain every property of the original file.
Privacy and Security Considerations
Before uploading sensitive media, check the verification service's current privacy terms and data-handling policy. A private recording, unreleased commercial video, or image containing personal information may deserve additional care.
Do not upload confidential material to an online service merely because the tool is convenient. For sensitive investigations, follow the organization's approved data-handling procedures and consider whether a local or controlled workflow is more appropriate.
Also remember that provenance signals do not determine copyright ownership, permission to publish, consent, or legal responsibility. Those issues require separate consideration.
Frequently Asked Questions
Can SynthID detect every AI-generated image?
No. It is designed to detect supported SynthID watermarks. Content generated by tools that do not use a compatible watermark may not be identified.
Does a negative result mean a video is real?
No. It means the supported signal was not detected in that check. It does not prove that the media was captured in the real world or that it has not been manipulated.
Can SynthID tell who created a file?
A watermark result does not automatically establish the identity of the person who generated, edited, or published the file. Attribution requires additional evidence.
Is a watermark the same as C2PA metadata?
No. Watermarking embeds a signal in content, while C2PA is a provenance framework that can record verifiable information about content and its history. They can complement one another.
Conclusion
Google's public expansion of SynthID Detector makes one form of AI-media verification more accessible. It gives users another way to investigate the origins of supported images, videos, and audio without relying entirely on visual guesses.
The practical rule is simple: treat a detected watermark as useful evidence about a supported signal, not as a complete judgment about the content. Treat a negative result just as carefully. The strongest verification combines provenance technology with source checking, file preservation, independent evidence, and a clear explanation of uncertainty.
