How to Check AI Images, Videos and Audio With Google's SynthID Detector

Google's AI detector is now open to everyone, but there is an important limitation

Google's SynthID Detector checks images, videos and audio for invisible watermarks added by supported AI systems. Visit synthid.com, upload a supported media file and review the result. A detected watermark indicates AI involvement, but no detected watermark does not prove that the content is authentic.

A photograph appears in a WhatsApp group claiming to show an explosion in Lagos. Someone shares an audio recording that supposedly captures a public official making a controversial statement. A convincing video begins circulating on TikTok, but nobody can identify where it originally came from.

Artificial intelligence makes these situations harder to assess because convincing images, recordings and videos no longer require a camera, microphone or the presence of the person being depicted.

On 7 October 2026, Google announced that its SynthID Detector was becoming publicly available worldwide in English, expanding a verification tool previously tested with journalists and media professionals.

The official announcement confirms that the detector can examine images, videos and audio created or edited using supported AI systems from Google and participating companies, including OpenAI, NVIDIA and Kakao. Apple support is expected later.

That expansion makes AI-origin verification more accessible. But it does not create a universal test for whether something online is real or fake. The distinction matters because an AI-generated image can lack a detectable watermark, while a genuine photograph can be shared alongside a completely false claim.

What is Google SynthID, and how does it detect AI-generated content?

SynthID is a digital watermarking technology developed by Google DeepMind. Rather than looking for unusual fingers, distorted faces or other visual clues commonly associated with AI images, it checks for signals embedded in content during generation or editing. These watermarks are designed to be invisible to the human eye or inaudible in recordings.

Google says the technology can withstand several ordinary modifications, including compression, resizing and certain image or audio adjustments. However, extensive modifications can make a watermark harder to detect.

This is different from many conventional AI detection tools, which attempt to estimate whether content resembles AI output.

SynthID's primary function is to identify a supported watermark. It does not simply look at a photograph and guess whether it appears artificial. That distinction also separates it from the AI-writing classifiers examined in TechView Africa's earlier report on why AI detectors sometimes flag human-written work. A watermark check and a statistical prediction about authorship are not interchangeable forms of evidence.

How to use Google's SynthID Detector

Google's public detector is available at synthid.com.

To check a suspicious image, video or audio file:

1. Open the official SynthID Detector website in your browser.
2. Follow any account or access instructions displayed.
3. Select the option to upload a supported image, video or audio file.
4. Allow the detector to examine the uploaded content for supported watermarks.
5. Read the result carefully, paying attention to whether a watermark was identified.
6. If the media concerns an important claim, investigate its source and context separately.

Where possible, use the original file rather than a heavily edited or repeatedly forwarded copy. An alternative is Google's Gemini app. Signed-in users can upload an image, video or audio file and ask whether it was created or edited using Google AI.

There is a significant difference between the two options. Google's current Gemini documentation says its SynthID verification recognises content from Google's own AI tools, whereas the new dedicated SynthID Detector supports participating partners.

Consequently, checking a file with Gemini should not be presented as equivalent to checking every participating company's content through the dedicated portal.

What does a SynthID result actually mean?

The most important part of using the detector is interpreting its findings correctly.

A detected watermark can provide meaningful evidence that AI was involved in producing or modifying content. It does not automatically mean every part of a photograph or video was fabricated.

For example, someone could edit a genuine photograph using a supported generative AI system. Detecting AI involvement would not, by itself, establish that the entire scene was invented.

Equally, a video without a recognised watermark could have been produced using an AI model that does not participate in SynthID's detection ecosystem.

This is why “no watermark detected” must never be translated into “verified as real.”

Can SynthID detect ChatGPT images and other AI-generated media?

Google lists OpenAI among the participants in its expanded detection ecosystem.

That is an important development, but it should not be interpreted as a guarantee that every image generated through ChatGPT or every output from any participating company's AI models will carry a detectable SynthID watermark.

Detection depends on the relevant product, model, watermarking implementation and the integrity of the media being examined.

Google's own models, including its image, video and audio generation systems, also use SynthID across supported outputs.

For users, the practical conclusion is straightforward: check the file, examine the result and avoid making assumptions solely from the company associated with the content.

How to verify a suspicious image when SynthID finds nothing

A negative watermark result should be the beginning of further checking, not the end.

TechView Africa recommends a four-part verification process for questionable images, videos and recordings.

First, investigate where the content came from. Look for its earliest identifiable publication and the original account or organisation responsible for sharing it. A forwarded WhatsApp image with no source deserves more scrutiny than material traceable to a credible firsthand publisher.

Second, conduct a reverse image search. Google's Lens image search can help identify other appearances of a picture. Google's About this image feature may also provide historical context where available.

A photograph presented as a fresh incident in Abuja might turn out to have appeared years earlier in another country. In that situation, the photograph could be genuine while the accompanying story is misleading.

Third, check available provenance information. Content Credentials, based on the C2PA standard, can provide signed information about a file's origin and editing history where compatible credentials exist. Their presence offers another useful signal; their absence is not proof of manipulation.

Finally, verify the underlying claim independently. If the media supposedly shows a government announcement, a bank's instruction or a public emergency, look for confirmation from the institution concerned and reliable independent reporting.

A believable face or familiar voice should not be enough to authorise a payment. TechView Africa's guide to verifying suspicious voice calls and deepfake emergencies explains why contacting the person through an independently trusted channel is more useful than relying on the recording itself.

Why this matters for Nigerian internet users

In Nigeria, suspicious media can circulate quickly through WhatsApp groups, social platforms and private messages, sometimes reaching people without a clear record of where the material originated.

The potential consequences extend beyond political misinformation. Fabricated photographs can be used in impersonation, fraudulent fundraising, fake product advertising and online harassment.

One especially sensitive example involves AI-generated intimate imagery. TechView Africa has previously explained what victims should do when threatened with fabricated explicit images.

For victims facing blackmail, watermark checking must not delay more important protective steps. Preserve evidence, avoid paying the perpetrator and report the abuse through appropriate channels. Do not upload private or intimate material to unfamiliar detection websites.

SynthID is useful because it gives ordinary users another way to investigate suspicious content. Its value is strongest when people understand its limits rather than expect it to deliver a final verdict.

Our Recommendation

Use the official SynthID Detector when you need to examine an image, video or audio file for supported AI watermarks. Treat a positive result as evidence of AI involvement, not proof that every detail in the content is false.

When no watermark is found, continue checking the original source, publication history, context and any available Content Credentials.

Above all, avoid sharing potentially damaging claims simply because an AI detector failed to identify a watermark. Whether media was generated by AI and whether the story attached to it is true are two different questions.