A public route to check for an embedded mark
Google has opened SynthID Detector beyond the media professionals who tested its early version. Its October 7 announcement says the tool is available globally in English and accepts images, video and audio.
Google describes support for content from its own AI systems and partners including OpenAI, NVIDIA and Kakao. Apple is listed as coming later. That is Google's account of coverage, not a claim that the tool recognises every output from every AI model.
The change gives ordinary readers a way to look for a technical clue that previously required more specialised access. It does not turn an upload into a complete fact-check.
The mark is inside the media
SynthID embeds a signal in generated content rather than relying only on a visible label or attached file information. Google says the signal is designed to survive common changes such as cropping or compression.
The design is useful when a file travels beyond the tool that made it. But resilience is not unlimited: Google's help documentation warns that repeated alterations can leave a watermark undetectable. Simple content and very small edits can also produce unclear results.
A positive result can indicate AI involvement. It cannot explain, on its own, whether a scene depicts a real event or whether a claim made in a clip is accurate. Those are different questions.
Do not merge two different checking routes
Google also provides verification within Gemini. Its help page says that route currently recognises SynthID marks from Google's AI tools, even though other companies have adopted the technology. That narrower description should not be confused with the standalone detector's announced partner coverage.
Gemini can separately inspect supported Content Credentials, which record information about origin and editing history. A watermark and a recorded history are complementary clues, not interchangeable guarantees.
The practical reading is cautious. No detected mark does not rule out another AI system or a mark lost through editing. Keep the original file and investigate its source before drawing a conclusion about an important image or recording.
What Google has announced, and what needs testing
Confirmed: Google links to a public detector. Official documentation explains watermarking and detection limitations. Model Current has not uploaded files or conducted an accuracy audit.
Company claims: the listed partner coverage and resistance to common edits. Open questions include performance across edited files and different creation tools.
The public portal is a new verification option. Its most useful role is to add evidence to a wider check, not replace the judgment that check requires.
Sources
- Google: public SynthID DetectorPrimary October 7 public-access announcement and named partner coverage, with Apple described as forthcoming.
- Google DeepMind: SynthIDPrimary technical overview of embedded watermarks, media types, editing resilience and public detector.
- Google Help: verification limitationsPrimary Gemini Help: Google-only SynthID recognition in Gemini, ambiguous results, alteration limits and distinct Content Credentials. Gemini-specific limits are not transferred to the standalone portal.
