SynthID Detector: a watermark check, not proof of truth
Google announced wider public access to SynthID Detector on 7 October 2026. It checks supported AI-generated images, videos and audio for SynthID watermarks. A missing watermark does not establish that a file is human-made, authentic or safe to trust.
Official publisher logo hosted by its publisher. Official source. Independent guide, not affiliated or sponsored. Artwork and trademarks belong to their owners.
Is the forwarded launch tip real?
Yes, with a wording correction. Google's official announcement dated 7 October 2026 says SynthID Detector is expanding to everyone, globally in English. It explicitly names Google and partners including OpenAI, NVIDIA and Kakao, with Apple coming soon. The announcement links directly to synthid.com.
This is not the first-ever launch. Google announced an earlier SynthID Detector portal on 20 May 2025 for media professionals and early testers. Today's news is expanded public access and partner coverage. “Soon” is not current Apple support.
Who is behind synthid.com?
The strongest source binding is Google's own official announcement linking that exact domain as its detector. On visiting the portal, its welcome notice also says the service is hosted by Google and governed by Google terms and privacy policy. We describe it as Google's officially linked portal; we did not independently audit domain-registration ownership or the operating entity's corporate records.
Use the exact official link rather than a similar-looking detection site. We inspected the public portal without accepting its terms, signing in or uploading a file.
What a watermark detector does
SynthID embeds an imperceptible signal into media at generation or publication time. The detector checks for supported signals already present in a file. The portal explicitly says it is not a general AI detector and only detects media from companies adopting the technology.
The new public announcement covers images, video and audio. SynthID also has a separate text-watermarking technology, but that does not mean the current public upload portal accepts arbitrary essays or detects every AI-written text.
Google's older overview still describes journalist testing. For current access, this guide follows the newer dated announcement and observed portal, rather than treating older wording as a current invitation-only restriction.
How to interpret results
| A supported watermark is found | This is evidence of a supported AI provenance signal. It does not tell you that every claim within the media is false, or explain every later edit. |
|---|---|
| No watermark is found | This is not proof of human creation or authenticity. The file could come from an unsupported tool, lack a supported signal or have changes that affect detectability. |
| The media was edited | Watermarks are designed to survive common transformations, not every possible modification. OpenAI's official help says more extensive changes can make a watermark undetectable. |
| A news claim looks believable | Check the source, date and context separately. Provenance is one clue, not a complete fact check. |
| Partner scope | Company participation does not mean every historical file, product, version or format from that company is supported. Follow current portal/model guidance. |
Before uploading a file
- Open the exact detector linked by Google's official announcement.
- Read the welcome terms and privacy notice. Decide whether to proceed; this guide has not accepted them on your behalf.
- Choose only media you have the right to upload. Avoid private conversations, medical records, identity documents and confidential workplace material.
- Use the highest-quality available original where possible, as the portal recommends. Do not interpret recompressed copies as an ideal detection test.
- Review the current sign-in, supported-format, size and usage requirements shown in your session. Exact file-size and daily numerical limits were not verified here.
- Upload only if you accept those conditions, then read the result and any uncertainty rather than reducing it to “real” or “fake”.
- For anything consequential, verify the original publisher and corroborating evidence independently before sharing.
These are editorial steps based on the public portal and official explanations, not a tested upload walkthrough.
Upload privacy is not zero retention
The observed portal notice says uploaded originals are processed in real time and deleted after results are returned. It separately says a digital signature of the media is tied to the user identifier for anti-abuse and daily quotas, then deleted after 24 hours. Third-party sign-in identifiers may also be retained for authentication and abuse prevention.
Do not simplify this to “nothing is stored” or “uploads are anonymous”. Those are the portal's stated terms, not an independent security audit. Check the current notice before choosing to upload.
Free access and limits
Google's announcement says public access is available globally in English. It does not give a subscription price. The observed welcome notice refers to daily quotas, but an exact free allowance, file-size ceiling or account entitlement was not verified without proceeding. This guide therefore does not promise unlimited free scans or quote an INR price.
OpenAI also operates a separate verification page for supported OpenAI images and audio. It looks for C2PA and SynthID signals. That separate page is not the same as Google's image/video/audio detector, and its formats and scope should not be mixed into this portal's limits.
Frequently asked questions
Does this detect all AI media?
No. It detects supported SynthID-watermarked media, not every AI output.
Does “no watermark” mean real?
No. An absent signal cannot establish authenticity or human authorship.
Are OpenAI and NVIDIA partnerships just a rumor?
No. They are named in Google's current official announcement. OpenAI also documents SynthID in its own provenance help.
Is Apple already supported?
The announcement says “soon, Apple”. Do not treat that as delivered support.
Can I upload text?
The current public announcement covers images, video and audio. Separate SynthID Text research is not proof of an arbitrary-text upload feature.
Sources: Google's 7 October 2026 expansion announcement, 20 May 2025 initial portal announcement, Officially linked detector and observed notice, DeepMind overview, OpenAI's separate verification portal, OpenAI provenance help. Checked 7 October 2026.