SynthID Detector Goes Public: What a Watermark Can Tell You
Google opened SynthID Detector globally in English on October 7. Understand supported media, watermark limits, and a practical verification workflow.
Bottom line
Google opened SynthID Detector globally in English on October 7. Understand supported media, watermark limits, and a practical verification workflow.
In this guide
Short answer
SynthID Detector is a useful provenance check for supported AI media, but it cannot establish that a claim is true. Google’s October 7 announcement expands public access. For editors, creators and small teams, the practical opportunity is adding one more evidence step before sharing unfamiliar media.
What Google announced
The launch post says the detector is available globally in English and accepts image, video and audio files. It names Google and partners including OpenAI, NVIDIA and Kakao; Apple is described as coming soon. That wording should not become a claim that every file from every partner is supported today.
Google’s technical overview explains that SynthID embeds an imperceptible watermark during generation. The detector looks for this signal rather than proving authorship from visual appearance alone. Google describes resilience to common edits, but DiscoverAI has not tested these claims across transformations or providers.
Three questions a result does not answer
Our editorial interpretation is that a provenance signal should be kept separate from factual verification. A watermarked image can illustrate a real event inaccurately. A genuine recording can be captioned with the wrong date. A synthetic diagram can accurately explain an idea.
Ask three independent questions: how was this file created, what does it claim, and what supports that claim? The detector contributes to the first question. Original reporting, source documents and corroboration address the others.
A result without a detected watermark should remain inconclusive. It could involve an unsupported tool, missing signal or a file outside the relevant detection path. Do not convert it into a human-made certificate or a reason to accuse someone of deception.
A practical editorial workflow
Keep the original file and the URL or sender from which you received it. Record the proposed caption before checking the media so the claim under review stays explicit. Submit only material you have permission to upload, especially when it contains private people or client information.
Save the result, date and file version. Then verify the event, location and quoted statement independently. If a colleague later crops or recompresses the file, preserve the relationship to the original rather than treating a new check as interchangeable evidence.
Use three internal outcomes: supported provenance signal, unresolved provenance, and independently verified factual claim. These labels are proposed editorial practice, not official SynthID output labels.
Try it before depending on it
For a proposed pilot, collect a small authorized set of known originals, supported AI outputs and edited copies. Have a colleague conceal their origins, then compare results against that answer key. Record inconclusive cases, upload failures and interpretation mistakes. This is a reader-run test; we have not measured accuracy.
A small team should adopt the detector when it improves its evidence trail without encouraging overconfidence. Keep human source review in the process. For a related everyday workflow, read [how to save AI answers as verified notes](/articles/save-ai-answers-as-verified-notes).
Transparency
How this guide was checked
Editorial accountability
Who checked this guide
- Evaluation type
- Research-based verification
- Last materially checked
- Evidence
- 2 listed sources
Hands-on testing is identified explicitly. Research-based coverage uses cited product documentation and other named sources; it does not imply every paid plan was used. Read the full methodology.
Editorial basis
What this guidance is based on
- Editorial basis
- Source-led analysis
- Primary references
- 2
- Products covered
- 1
- Last checked
- 2026-10-08
Important limits
- • DiscoverAI has not independently reproduced research findings or performed the proposed product tests.
- • Availability, policies, billing and feature entitlements may change.
Sources and verification
Product details and claims were checked against the following primary sources.
Frequently asked questions
Is SynthID Detector publicly available?
Google says global English access began October 7, 2026; check current file and account requirements.
Does no watermark mean human-made?
No. A missing detected signal does not rule out AI generation.
Does a watermark prove a claim is true?
No. Provenance and factual accuracy require separate checks.
Did DiscoverAI test detection accuracy?
No. This article analyzes official documentation and proposes a pilot.
Tools mentioned in this article
Google Gemini
Google's deeply integrated AI assistant with unmatched access to Google's ecosystem
Gemini combines powerful AI with Google's vast data ecosystem — Search, Gmail, Docs, YouTube, and more — for a uniquely integrated experience.
Read next
