Google SynthID Detector
Quick Take
SynthID Detector answers one narrow question well: does this file carry a SynthID watermark? Google DeepMind has used SynthID since 2023 to embed an invisible signal in images, video, and audio made with its models, and Google says it has watermarked more than 180 billion images and videos. Since October 7, 2026, anyone can upload a file to the detector portal and check. It is useful for journalists, editors, and trust and safety teams, as long as you understand what a “no” result does and does not mean.
Watermark Detection Is Not AI Detection
These two things get confused constantly, and the difference decides how much a result is worth.
Watermark detection looks for a specific signal that a generator deliberately embedded at creation time. If the signal is there, you have strong evidence about where the content came from. SynthID Detector does this, and only for SynthID.
General AI detection tries to guess whether arbitrary content was made by AI by looking for statistical patterns. Those tools work on content from any source, but they make mistakes in both directions, as our guide to whether AI detectors can be trusted explains.
So a positive SynthID result is meaningful: some or all of the file came from a model that uses SynthID. A negative result is not proof of anything. The content may have come from an AI tool that does not use SynthID, the watermark may have been damaged by heavy editing, or a person may have made it. Google’s own help pages say a missing watermark does not confirm a file is authentic.
What It Covers
Media types. Images, video, and audio. SynthID also watermarks text from the Gemini app, but the detector portal Google opened in October 2026 is described for media files.
Companies. Content made with Google’s AI tools, plus partners that have adopted SynthID. Google’s October announcement names OpenAI, NVIDIA, Kakao, and Apple, and OpenAI’s own help center confirms that images made with ChatGPT, Codex, and its API carry SynthID alongside Content Credentials; Google DeepMind’s SynthID page still describes Apple’s support as upcoming, so check before relying on that one. Content from other companies’ AI tools is outside its scope unless they adopt SynthID.
Durability. Google says the image and video watermark is designed to survive cropping, filters, frame-rate changes, and lossy compression, and the audio watermark to survive added noise, MP3 compression, and speed changes. Its help pages also say detection may fail after many alterations.
Where to Check
- The SynthID Detector portal, open to everyone globally in English since October 7, 2026. You upload a file and it reports whether it found a SynthID watermark. Google’s announcement does not list file limits, describe the result screen in detail, or say how uploaded files are handled, so avoid uploading anything confidential.
- The Gemini app. Upload a file and ask, for example, “Was this made with Google AI?” Google documents limits here: one file at a time, up to 100 MB, videos under 90 seconds, audio under an hour, and roughly ten checks per media type in a rolling 24 hours. You need to be signed in. Google’s help pages describe this check for watermarks from Google’s own AI tools.
- Google Search and Chrome now include similar verification, which Google says handles more than a million requests a day across its surfaces.
The Gemini app also reads Content Credentials, the C2PA provenance record some cameras and editing tools attach to files, which can show a file’s origin and edit history whether or not AI was involved. That is a different signal from SynthID and can be stripped when a file is re-saved or screenshotted. Our AI watermark glossary entry explains how the two differ.
Reading a Result
Based on Google’s documentation for the Gemini app check:
- Watermark found means all or part of the file was created or edited with a model that uses SynthID. For video, Gemini says which portions had a watermark detected. A partial result can mean someone edited a real photo with an AI tool, or spliced AI footage into real footage.
- No watermark found means the file was not created or edited with a SynthID model, as far as the check can tell. It does not mean the file is real.
- Inconclusive results happen, for example when an image is too small or an AI edit was too slight to carry a readable watermark.
Who Should Use It
Journalists and fact-checkers can quickly confirm whether a viral image or clip came from a Google or partner model. Editors and media reviewers can check submitted images before publication. Trust and safety teams can add it as one signal in a review process. Creative teams can confirm what their own AI-assisted assets will show when someone checks them.
In each case, treat it as one piece of evidence alongside source checks, reverse image search, metadata, and contacting the person who shared the file.
Limitations
- One watermark system only. It cannot see content from AI tools that do not use SynthID.
- “No watermark” proves nothing about whether a person made the content.
- Heavy editing can defeat it. Google says detection may fail after many alterations.
- English only for the portal, according to Google’s announcement.
- Little published detail on the portal’s file limits, confidence wording, or how uploads are stored.
- Screenshots and crops weaken it. Google’s own advice is to crop tightly around an image and not to check collages.
Related Pages
- Guide: Can AI detectors be trusted?, on the difference between detection, watermarking, and provenance
- Glossary: AI watermark, covering SynthID-style watermarks and C2PA Content Credentials
- Google Flow, Veo, and Gemini, Google tools whose output carries SynthID
- Google Genie 4, whose video and audio are also watermarked with SynthID
Continue learning
Explore related guides, tools, workflows, and prompts that help you go deeper into this topic.
See how this tool fits into a workflow
Browse step-by-step AI workflows that use ChatGPT, Claude, Gemini, and other tools.
Frequently Asked Questions
What does SynthID Detector do?
It checks whether an image, video, or audio file contains a SynthID watermark, an invisible signal that Google embeds in content made with its AI models and that partners including OpenAI, NVIDIA, and Kakao have adopted. If it finds one, the file was created or edited, at least in part, by a model that uses SynthID.
Can SynthID Detector tell me if any image is AI-generated?
No. It only finds SynthID watermarks. Content from AI tools that do not use SynthID, or content whose watermark was damaged by heavy editing, will show no watermark. That result does not mean a person made it, and Google says a missing watermark does not confirm a file is authentic.
Who can use SynthID Detector?
Since October 7, 2026, anyone, globally, in English, according to Google. It started in 2025 as a portal for early testers such as journalists and researchers. You can also ask the Gemini app whether a file was made with Google AI, and Google has added similar checks to Search and Chrome.
Does SynthID Detector check text?
Google's October 2026 announcement covers images, video, and audio. SynthID also watermarks text from the Gemini app, but the public detector portal is described for media files, not text, and other companies' chatbots generally use different methods or none.
Which companies' content can it detect?
Content made with Google's AI tools, plus partners that have adopted SynthID. Google's October 2026 announcement names OpenAI, NVIDIA, Kakao, and Apple, while Google DeepMind's SynthID page still describes Apple's support as upcoming. Content from any other company is outside its scope unless that company has adopted SynthID.
Last updated: