Google SynthID Detector: How to Check if an Image, Video, or Audio Is AI-Generated

AI-generated images, videos, and audio are becoming increasingly realistic, making it difficult to determine where digital content comes from. 

Google SynthID Detector helps address this problem by checking media for invisible digital watermarks associated with supported AI-generation systems.

The tool is now available globally in English and supports checking images, videos, and audio. Google announced its wider public availability on October 7, 2026, extending access beyond its initial testing with media professionals. 

However, SynthID Detector does not identify every AI-generated file. It checks for compatible watermarks, so a negative result does not necessarily mean that content was created by a human.

What Is Google SynthID Detector?

Google SynthID Detector is a media verification tool that checks whether an image, video, or audio file contains an invisible SynthID watermark. Google DeepMind developed SynthID to help identify content generated or edited using supported AI systems.

Unlike a visible watermark containing a logo or text, a SynthID watermark is embedded directly into digital content and cannot ordinarily be seen or heard. Its purpose is to provide a signal that detection technology can recognize without noticeably affecting the media's quality.

Google introduced SynthID in 2023 and subsequently expanded the technology across images, videos, audio, and text. 

The public detector's October 2026 expansion includes content created using Google's AI tools and supported partner systems, including OpenAI, NVIDIA, and Kakao. Apple was identified as a future addition in Google's announcement.

SynthID is useful for checking the provenance of digital media, but it should not be confused with a universal AI detector. 

Its results depend on whether the content carries a watermark the system can recognize. Google DeepMind explains the underlying technology and its supported applications.

How Does Google SynthID Detector Work?

SynthID embeds an imperceptible watermark into AI-generated content during the generation process. 

The detector then analyzes the uploaded media to determine whether it contains a recognizable watermark pattern.

The process has three main stages:

1. Watermark embedding: A supported AI system incorporates an invisible signal into generated content.
2. Media analysis: The detector examines the uploaded file for the relevant watermark.
3. Result interpretation: The tool reports whether it detects a watermark and, where supported, identifies portions of the media associated with it.

The watermark is designed to remain detectable after common transformations, such as resizing, recoloring, compression, or changes to video frame rates. However, repeated or extensive alterations can weaken the signal and make detection more difficult.

For video and audio, the analysis can also identify specific portions associated with a watermark. 

This is useful when a file contains a mixture of generated and other material. A detected watermark in one segment does not necessarily mean that every frame or sound in the entire file was AI-generated.

The essential distinction is that SynthID checks for a specific embedded signal rather than judging whether media simply looks or sounds artificial.

Professionals looking to apply AI in business can explore the AI Agents for business course from Texas McCombs. This 13-week online program covers AI agents, prompt engineering, RAG, and no-code AI workflows for business automation.

How to Check if an Image, Video, or Audio Is AI-Generated Using SynthID Detector

You can check supported media through Google's SynthID verification experience. Google also provides verification through Gemini, where you can upload a file and ask whether Google AI created or edited it.

Step 1: Open the verification tool

Visit the Google DeepMind SynthID page to access the available verification options. Alternatively, open Gemini and sign in to your account.

Step 2: Upload the media file

In Gemini, select the option to add a file and upload the image, video, or audio recording you want to examine. You can also use supported Google Photos or Google Drive upload options when available.

For Gemini verification, Google currently specifies a maximum file size of 100 MB. Videos must be shorter than 90 seconds, and audio files must be shorter than one hour. 

Only one media file can be checked at a time. These limits apply to the Gemini workflow and should not be assumed to be identical for every detector interface.

Step 3: Ask a direct verification question

After uploading the file, ask a focused question such as:

• “Was this image created or edited using Google AI?”
• “Does this video contain a SynthID watermark?”
• “Was this audio generated using Google AI?”

A specific question helps clarify whether you are investigating AI generation, editing, or the presence of a watermark.

Step 4: Review the result

Gemini returns a response based on the available verification signals. Depending on the content and result, it may report a detected watermark, no detected watermark, or insufficient evidence to determine the origin confidently.

For videos, the response can identify portions where SynthID was detected. Read the explanation carefully instead of assuming that the result applies uniformly to the entire file.

Step 5: Interpret the finding in context

Treat the result as one piece of evidence. If the content is being used to support a news report, research finding, public allegation, or other consequential claim, verify its source and context independently.

A watermark can provide useful evidence about AI involvement, but it cannot establish whether a depicted event actually happened or whether the accompanying claims are truthful.

How to Interpret SynthID Detector Results

A SynthID result should be interpreted according to what the tool actually detected—not as a definitive judgment about whether the entire file is authentic.

Google's Gemini guidance explains that a detected SynthID watermark indicates that all or part of an image or video was created or edited using Google's AI models. 

It also notes that some media may not contain enough information to watermark and that minor edits may be too small to detect reliably.

Consider an image of a city skyline. If SynthID detects a watermark, the finding supports the conclusion that a supported AI workflow contributed to the image. 

It does not establish that the entire scene is fictional: the image could combine a real photograph with AI-generated elements.

Similarly, if no watermark is detected in a video, you cannot conclude that the footage is authentic. It may have been generated using an unsupported system, edited in ways that weaken the signal, or produced without a compatible watermark.

The correct interpretation is therefore narrow: SynthID provides evidence of supported AI involvement, not a universal verdict on authenticity.

What Are the Limitations of Google SynthID Detector?

SynthID improves transparency around AI-generated media, but several limitations affect how its results should be used.

It does not detect every AI system. The detector looks for supported watermarks. Media produced by an AI tool that does not embed a compatible watermark may not be identified.

A negative result is not proof of human creation. The absence of a detected watermark only means that the system did not find a recognizable signal. It does not establish the media's original source.

Extensive editing can affect detection. SynthID is designed to withstand common transformations, including compression and resizing. However, Google acknowledges that repeated alterations can make a watermark harder to detect.

Minor AI edits may be difficult to identify. A small change to an existing image or video may not contain enough detectable watermark information to produce a clear result.

A watermark does not prove that a claim is true. Even when a watermark is detected, it does not establish whether the media depicts a real event, whether its caption is accurate, or whether the content has been presented honestly.

Verification signals have different purposes. SynthID identifies embedded watermarks, while Content Credentials can provide information about a file's origin and editing history. Neither should be treated as a complete substitute for examining the original source and surrounding evidence.

These limitations make SynthID most useful as one component of a broader media-verification process, particularly when authenticity matters.

For developers interested in building AI-powered applications, the full stack development course​ from Texas McCombs offers a 14-week online learning experience covering full-stack development, LLM integration, AI agents, APIs, and AWS deployment.

Conclusion

Google SynthID Detector offers a practical way to check images, videos, and audio for invisible watermarks associated with supported AI-generation systems. 

Its ability to identify watermarked content can help users investigate media provenance without relying exclusively on visual clues or file metadata.

However, the tool is not a universal AI detector, and a missing watermark does not prove that content is human-created. 

For reliable verification, interpret the result alongside the original source, available provenance information, and independent evidence.

Frequently Asked Questions

1. What is Google SynthID Detector?

Google SynthID Detector checks images, videos, and audio for invisible SynthID watermarks associated with supported AI-generation systems.

2. How do I check if an image is AI-generated using SynthID?

Open Google's SynthID verification experience or Gemini, upload the image, and ask whether it was created or edited using Google AI.

3. Can SynthID Detector check videos and audio?

Yes. The public detector supports images, videos, and audio. Available file limits and verification features may vary by interface.

4. Does SynthID detect content created by ChatGPT or other AI tools?

Google's October 2026 announcement includes supported partner systems such as OpenAI, NVIDIA, and Kakao. Detection depends on compatible watermarking.

5. What does it mean if SynthID does not detect a watermark?

It means no recognizable watermark was detected. The media could still be AI-generated by an unsupported system or contain a watermark that cannot be reliably identified.

6. Can SynthID detect AI-edited images?

It can identify supported AI-generated or edited content when a detectable watermark is present. Minor edits may not always produce a detectable signal.

7. Is Google SynthID Detector free to use?

Google announced wider public access to SynthID Detector in English. Check the official verification interface for current access requirements and usage limits.

8. Is SynthID Detector 100% accurate?

No detector should be treated as infallible. SynthID provides evidence of supported watermarking, but it cannot identify every AI-generated file or independently prove authenticity.

As AI becomes more integrated into digital products and software workflows, developers also need to understand how to build AI-powered applications responsibly. Those looking to develop these skills can explore Generative AI and AI agents for software development.