Figuring out whether a suspicious image, video, or audio clip was made with AI is about to get a little easier. Google DeepMind is opening its SynthID Detector to everyone, turning what was previously a limited verification tool for media professionals into a publicly available way to check AI-generated content.

The expansion arrives as generative AI makes it increasingly difficult to tell where digital media came from simply by looking at it. Google says the SynthID Detector is now available globally in English, allowing anyone to upload supported media and check whether it carries a SynthID watermark.

There is an important catch, though: this isn’t a universal “AI detector” that can identify anything made by any AI system. It is primarily looking for an invisible SynthID watermark embedded in supported AI-generated content.

SynthID Detector is now open to everyone

Google first introduced SynthID in 2023 as a way to watermark AI-generated content without visibly changing it. The technology has since expanded beyond images to cover video and audio, and Google has also developed SynthID watermarking for text.

The SynthID Detector launched in 2025 as a verification portal, but access was initially aimed at journalists, media professionals and researchers. After testing the system and collecting feedback, Google is now making the detector available to everyone.

You can upload an image, video or audio file to the detector and have it checked for a SynthID watermark. If the watermark is found, the system can help identify portions of the media that are likely to contain the watermark.

Google says the expanded detector is available globally in English starting October 7.

It can detect more than Google’s own AI

This is where the announcement gets particularly interesting.

SynthID is no longer limited to content generated by Google’s own AI systems. Google says the expanded detector can identify SynthID watermarks in AI-generated images, video and audio from Google and participating partners, including OpenAI, NVIDIA and Kakao. Apple is also expected to join the ecosystem soon.

That matters because watermarking becomes considerably more useful when it isn’t tied to one company’s AI products.

The broader idea is straightforward: instead of trying to guess whether something “looks like AI,” a verification system can look for a signal deliberately embedded when the content was generated.

That doesn’t make AI-generated media impossible to disguise, but it gives people another piece of evidence when trying to determine where something came from.

How SynthID actually works

SynthID doesn’t put an obvious label over an AI-generated image or add a watermark that you can see in a corner. Instead, it embeds an imperceptible digital signal into the content.

For images and video, the watermark is designed to remain detectable even after common changes such as cropping, filters, changes to frame rates and lossy compression. For audio, Google says SynthID is designed to remain detectable after modifications such as noise, MP3 compression and changes in playback speed.

That makes the approach different from ordinary metadata.

Metadata can tell you things about a file’s origin, but it can also disappear when a file is edited, exported or uploaded to another service. SynthID is embedded into the content itself, giving it a different layer of persistence.

Google has also expanded SynthID to text generated by Gemini. In that case, the system subtly adjusts the probability of token choices during generation to create a detectable statistical pattern. Google says the text watermark works best with longer pieces of generated text.

But there’s an important distinction for anyone using the new public detector: Google’s October announcement specifically describes the expanded SynthID Detector as checking image, video and audio files. It shouldn’t be treated as a general-purpose detector for every piece of AI-written text.

SynthID isn’t a magic AI detector

This is probably the most important thing to understand about the announcement.

If you upload an image and SynthID doesn’t find a watermark, that does not automatically prove the image was created by a human. It means the detector didn’t find the SynthID signal it is looking for.

That’s a fundamentally different claim.

SynthID works because participating AI systems put the watermark into their generated content. An image created with an unrelated AI system that doesn’t use SynthID won’t suddenly acquire a SynthID watermark just because it was generated by AI.

Even Google has previously described SynthID as an imperfect technology and positioned it as one part of a broader set of approaches for identifying digital content.

So the best way to think about SynthID Detector is as a provenance and verification tool rather than an all-knowing AI-content judge.

If it finds a SynthID watermark, that’s useful evidence that the content was generated or altered by a participating AI system. If it doesn’t find one, there may simply be no SynthID watermark to detect.

Google is building verification into more products

The standalone detector isn’t Google’s only way of checking AI-generated media.

Google has already added SynthID verification capabilities to Search, the Gemini app and Chrome. The company says those built-in verification features now handle more than 1 million requests per day.

That could ultimately be more important than the standalone detector itself.

Most people aren’t going to download a suspicious image, open a dedicated verification website and investigate its provenance every time they encounter something strange online. Verification becomes much more useful when it appears directly inside the tools people already use to browse the web.

Google has also been working with other industry players on content provenance and transparency, including C2PA Content Credentials, which can provide information about how digital content was created and modified.

The long-term goal is therefore bigger than SynthID alone: give people multiple signals that help establish where digital content came from and how it has been changed.

A small but useful weapon against the AI-content problem

The timing makes sense. Generative AI has made producing convincing images, videos and audio dramatically easier, while social platforms make it just as easy for that material to spread without its original context.

A watermark can’t solve that problem by itself. It can’t tell you whether a claim accompanying an AI-generated video is true, and it can’t identify every synthetic file floating around the internet.

What it can do is provide something much more concrete than asking whether an image “looks AI-generated.”

Google’s decision to open SynthID Detector to everyone also marks a shift in who gets access to that kind of verification. Instead of leaving provenance checks primarily in the hands of journalists and researchers, Google is putting the tool directly in front of ordinary users.

And as more AI companies adopt interoperable ways of identifying generated media, that invisible watermark could become a useful piece of infrastructure for the increasingly synthetic internet.

For now, though, the rule is simple: a SynthID match is meaningful evidence; a lack of a match is not proof that something is human-made. That’s a much more realistic way to use Google’s new detector than treating it as an AI-content lie detector.