Google SynthID is a digital watermarking technology from Google DeepMind designed to help identify content generated or altered with AI. Instead of placing a visible label or changing the appearance of a file, SynthID embeds an imperceptible signal directly into AI-generated images, videos, audio, and text.

The technology was first introduced in 2023 for AI-generated images. Since then, Google has expanded SynthID to other types of media and integrated it into products such as Gemini, Imagen, Veo, and Lyria. In October 2026, Google also opened its SynthID Detector to everyone globally in English, allowing people to check images, videos, and audio for supported SynthID watermarks.

The basic idea is fairly simple: rather than trying to guess whether something “looks like AI,” SynthID looks for a signal that was embedded when the AI-generated content was created.

That distinction is important.

How does Google SynthID work?

SynthID works by embedding a digital watermark into AI-generated content. The watermark is designed to be invisible or inaudible to people while remaining detectable by software.

For images and video, the watermark is incorporated into the content itself. Google says the system is designed to maintain the quality of the generated media while remaining detectable after common modifications such as cropping, applying filters, changing colors, altering video frame rates, and using lossy compression.

For audio, SynthID uses a different technique. Google describes converting audio into a representation of its frequency content and embedding the watermark into that signal. The resulting watermark is inaudible to people and is designed to remain detectable after common changes such as adding noise, MP3 compression, or changing playback speed.

Text works differently again. Large language models generate text one token at a time, with each possible token receiving a probability score. SynthID can subtly adjust those probabilities so that the resulting text contains a statistical pattern that a detector can identify. The changes are intended to be invisible to the reader and not noticeably affect the quality of the writing.

So, despite being called a watermark, SynthID isn’t a tiny visible logo hiding somewhere in an image. Think of it more like a machine-readable fingerprint embedded into the generated content.

What types of content can SynthID watermark?

SynthID now covers four major types of AI-generated content:

• Images
• Video
• Audio
• Text

Google’s current SynthID technology page says the watermark can be embedded into AI-generated images, audio, text, and video.

Google has progressively incorporated the technology into its generative AI ecosystem. That includes image generation through Imagen, video generated by Veo, audio generated by Lyria, and AI-generated text from Gemini.

Google has also expanded SynthID beyond its own products through partnerships with other AI companies. As of October 2026, Google says the SynthID ecosystem includes OpenAI, NVIDIA, and Kakao, with Apple support expected soon.

OpenAI separately confirms that images generated with ChatGPT, Codex, or the OpenAI API can include SynthID watermarks alongside Content Credentials, while audio generated through supported OpenAI tools can also use SynthID.

That is an important development because it means SynthID is increasingly being used as an industry-wide provenance signal rather than simply a Google-only technology.

What is the SynthID watermark actually for?

The primary purpose is content provenance.

As generative AI becomes capable of producing increasingly realistic photographs, videos, voices, music, and written content, it can become difficult to determine how something was created simply by looking or listening to it.

SynthID provides another piece of evidence.

For example, imagine someone posts a photorealistic image of a person doing something that never actually happened. If the image was generated using an AI system that embeds a SynthID watermark, a compatible detector may be able to identify that watermark and indicate that the content was generated or altered with supported AI technology.

This can be useful for journalists, researchers, platforms, educators, social-media users, and ordinary internet users trying to understand where digital media came from.

Google says its goal is to provide more context about AI-generated content and help reduce problems such as misinformation and misattribution.

Is SynthID an AI detector?

Not exactly.

This is probably the most important distinction to understand.

A conventional AI detector might examine a piece of content and attempt to determine whether it was generated by AI based on characteristics it finds in the content.

SynthID takes a different approach when a supported watermark is present: it looks for the watermark.

In other words, SynthID is fundamentally a watermarking and verification system, not a universal “AI detector” that can look at any file and magically determine whether AI was involved.

If you upload a photograph that was created by an AI system that doesn’t use SynthID, the absence of a SynthID watermark does not prove that the photograph was made by a human.

Likewise, if a watermark has been damaged or removed, a detector may not be able to identify it.

Google itself has acknowledged that SynthID isn’t foolproof, particularly against extreme manipulations. Its original image research described the technology as robust against many common modifications but not immune to every possible attack or transformation.

That’s why a SynthID result should be treated as a provenance signal rather than an absolute verdict about the history of a piece of content.

How to check an image, video, or audio file with SynthID Detector

Google introduced SynthID Detector in 2025 as a verification portal initially aimed at journalists, media professionals, and researchers. In October 2026, Google expanded access globally in English so anyone can use it.

The process is straightforward.

1. Open the SynthID Detector website.
2. Upload a supported image, video, or audio file.
3. Let the detector scan the file for a SynthID watermark.
4. Review the result to see whether a SynthID watermark was detected and, where applicable, which portions of the content are more likely to contain the watermark.

Google says the detector can identify watermarked portions rather than simply returning a single yes-or-no result. For example, it can highlight areas of an image or specific segments of audio where a watermark is more likely to be present.

The public detector currently focuses on images, videos, and audio. Although SynthID also supports text watermarking, Google’s October 2026 announcement specifically describes the public SynthID Detector as checking image, video, and audio files.

You can also check SynthID through Gemini

You don’t necessarily have to visit the standalone detector every time.

Google has built SynthID verification into Gemini and other Google products. Google says you can upload an image, video, or audio clip to Gemini and ask whether it was created or altered using Google AI. Gemini can check for a SynthID watermark and provide the result.

Google has also expanded verification capabilities into Search and Chrome.

This matters because provenance checking becomes much more useful when it is integrated into the places where people already encounter content. Google says its verification features across Search, Gemini, and Chrome now handle more than 1 million requests per day.

What happens if SynthID finds a watermark?

A detected watermark indicates that the content contains a SynthID signal associated with a supported AI-generation workflow.

That does not necessarily tell you everything about the content.

For example, it may indicate that an image was generated or edited using a supported AI system, but the watermark alone does not establish who created the content, who owns it, whether the accompanying claim is true, or how much of the final work was produced by AI.

This is one reason Google describes SynthID as part of a broader content-transparency effort rather than a complete solution to online misinformation.

Content provenance can involve multiple signals, including metadata, digital signatures, Content Credentials, and watermarks. Google is also involved in the Coalition for Content Provenance and Authenticity (C2PA), an industry effort focused on documenting the origin and editing history of digital content.

What if SynthID doesn’t find a watermark?

This is where you need to be careful.

“No SynthID watermark detected” does not mean “this content was made by a human.”

It could simply mean the content was never watermarked with SynthID.

For example, an image generated by an AI model that does not use SynthID would not necessarily contain a SynthID watermark. A watermark can also potentially become undetectable after sufficiently aggressive manipulation.

OpenAI makes a similar point about provenance systems: if a watermark or metadata isn’t detected, that does not necessarily prove that content was not generated by AI because provenance signals can sometimes be stripped or lost.

So the correct way to interpret SynthID is:

Watermark detected: There is evidence that supported AI-generated content is present.
Watermark not detected: SynthID did not find its signal. That is not proof of human authorship.

That distinction is easy to miss, but it is critical.

How much content has Google watermarked with SynthID?

SynthID has grown dramatically since its introduction in 2023.

When Google introduced SynthID Detector in May 2025, it said more than 10 billion pieces of content had already been watermarked.

By October 2026, Google said the technology had been used to watermark more than 180 billion images and videos, along with audio equivalent to 240,000 years of content.

That enormous increase reflects how deeply watermarking has become integrated into Google’s generative AI products and services.

It also explains why a verification system becomes increasingly useful as AI-generated media spreads across the internet.

Can SynthID be removed?

SynthID is specifically designed to survive many ordinary modifications, but that doesn’t mean the watermark is indestructible.

Google’s original research found that the image watermark could remain detectable after common transformations such as filtering, color changes, brightness adjustments, and lossy compression.

For audio, Google similarly designed the watermark to withstand common changes such as noise, compression, and speed adjustments.

However, Google does not claim that SynthID can survive every possible manipulation. Deliberate or extreme modifications can make detection more difficult.

This is another reason watermarking should be considered one layer of content provenance rather than a perfect anti-deepfake mechanism.

SynthID vs. metadata and C2PA

SynthID and metadata solve related but different problems.

Traditional metadata can contain information about a file’s creator, software, creation date, editing history, or other attributes. The problem is that metadata can be removed, modified, or lost when a file is processed or uploaded to another service.

SynthID instead embeds a signal into the content itself.

That makes the approaches complementary.

Google’s original SynthID research specifically noted that the watermark can remain detectable when metadata has been removed, because the signal is embedded in the image rather than stored solely as file metadata.

C2PA Content Credentials take yet another approach by creating a standardized, cryptographically verifiable record of a piece of content’s origin and editing history.

The broader industry is increasingly combining these techniques rather than treating them as competing technologies. OpenAI, for example, says its generated images can contain both C2PA Content Credentials and SynthID.

Why does Google SynthID matter?

AI-generated content is becoming so realistic that visual inspection alone is becoming an increasingly unreliable way to determine how something was made.

A watermark doesn’t solve that problem by itself, but it provides something much more useful than simply asking an AI model, “Does this look AI-generated?”

If the generation system inserted a durable provenance signal at creation time, a compatible verification tool has something concrete to look for.

That’s particularly useful for newsrooms investigating images and videos, researchers studying synthetic media, platforms trying to provide users with more context, and ordinary people who encounter suspicious content online.

The bigger idea behind SynthID is therefore not simply “detect AI.”

It is to make AI-generated content carry a machine-readable clue about its origin.

And with Google now opening SynthID Detector to the public and expanding the technology to content created by partners such as OpenAI, NVIDIA, and Kakao, that clue is becoming part of a broader effort to make the origins of digital media easier to verify.

The bottom line

Google SynthID is Google’s digital watermarking technology for identifying AI-generated and AI-altered content. It embeds imperceptible signals into images, video, audio, and text so compatible systems can later check whether those signals are present.

The technology began with AI-generated images in 2023 and has since expanded across Google’s generative AI ecosystem and into partnerships with other AI companies. Google now offers SynthID verification through Gemini, Search, and Chrome, while the standalone SynthID Detector is publicly available globally in English for checking images, videos, and audio.

But SynthID isn’t a magic AI detector. A detected watermark can provide useful evidence that supported AI generation or editing was involved, while the absence of a watermark does not prove that a human created the content.

As synthetic media becomes increasingly difficult to distinguish from conventional photography, recordings, and writing, technologies such as SynthID are becoming an important part of the infrastructure for figuring out where digital content came from.