SynthID Detector Opens to Everyone with Partner Support

Introduction

Google announced on October 7, 2026 that it is expanding SynthID Detector, a tool for checking whether media was made with AI, to everyone. In its blog post, Google says the detector is "available globally in English starting today." Last year, Google writes, it introduced an early version of the detector to help media professionals; the new release opens it to the general public through a dedicated website, SynthID.com, according to Ars Technica.

The update matters because the tool no longer checks only Google's own watermark. Google says anyone can now check whether an image, video or audio file was made with AI from Google or from partners "including OpenAI, NVIDIA, Kakao, and soon, Apple."

Feature Overview

1. A public website for verification

Ars Technica reports that the detector now lives on SynthID.com. It notes that people outside the trusted-tester program could previously check content only by asking Gemini. Google says SynthID launched in 2023 and that it has watermarked "over 180 billion images and videos, along with 240,000 years of audio content."

2. Support for partner watermarks

Ars Technica reports that the earlier detector checked only Google's own SynthID, meaning a ChatGPT image carrying SynthID would not have been flagged, and that partners' distinct watermarks are now supported. Google's post names OpenAI, NVIDIA and Kakao as partners, with Apple to follow.

3. Broad file-format coverage

TechCrunch lists the supported formats as follows.

Text watermark detection is not mentioned in this launch, so the site should not be assumed to check written content.

4. Account login and a usage quota

Ars Technica reports that the site requires login and supports Google, OpenAI and Apple accounts. Google says each user gets "a daily usage quota of approximately 10 image, video, and audio checks." Google engineers told Ars the limit is meant to prevent people from using repeated checks to develop SynthID removal tools.

5. Simpler results than the internal tool

According to Ars Technica, the website gives less detail than Google's internal tool, which highlighted regions containing SynthID pixels. The public site only says whether SynthID was used to produce or edit the media.

6. Links to built-in verification

Google says the detector joins built-in verification in Search, the Gemini app and Chrome, which it says "now regularly handle over 1 million requests daily." TechCrunch reports that Google's Nano Banana, Veo and Lyria models, and tools including Gemini, Flow, ProducerAI and Vids, watermark with SynthID.

Usability Analysis

For a journalist, teacher or ordinary user who receives a suspicious image, the workflow is short: sign in, upload a file, and read whether SynthID is present. A single destination that understands several vendors' watermarks removes the need to know which company's checker to use. TechCrunch notes that OpenAI maintains its own checking site at openai.com/research/verify, and that Microsoft and Meta have their own watermark standards, which TechCrunch, citing a Reuters report, says often fail to identify content from their own models.

The practical limits are clear. A result tells you whether a watermark is present, not whether a piece of media is truthful. A clean result does not prove that content is human-made. Ars Technica notes that Meta uses different watermarking technology, some providers use none, and open-weight models run locally produce content without labels. According to Ars, the site will carry a disclaimer saying so. The roughly 10 checks per day is enough for casual verification but not for newsrooms screening large volumes.

Pros and Cons

Strengths

• One public site covers images, video and audio in many common formats.
• Partner support extends checks beyond Google's own models.
• Google gives a specific scale for the watermarking effort, with more than 180 billion images and videos marked.

Limitations

• Content from unwatermarked sources, other watermark systems or local open-weight models cannot be flagged.
• Login and a daily quota of about 10 checks restrict heavy use.
• The public result is less detailed than the internal tool's region highlighting.
• English-only availability at launch.

Outlook

Ars Technica suggests that C2PA labeling of real media, which Pixel phones already do, is the complementary approach: watermarks identify synthetic content, while provenance labels vouch for authentic capture. The two methods answer different questions, and neither covers every case alone. Whether other major vendors adopt SynthID, or whether the planned Apple support arrives on a particular schedule, is not stated in the sources reviewed, so those remain open points. The detector's value will grow with the number of generators that embed compatible watermarks.

Conclusion

SynthID Detector's public launch is a practical step for anyone who needs a quick answer on whether media came from a participating AI system. Its limits are equally practical: it checks only for watermarks, it depends on partner adoption, and the quota restricts volume. It suits journalists, educators and general users doing occasional checks, and it works best alongside provenance labels and critical reading. We rate it 4 out of 5.

Editor's Verdict

SynthID Detector Opens to Everyone with Partner Support earns a solid recommendation within the Gemini space.

The strongest case for paying attention: a single public site checks images, video and audio across many common file formats. That alone raises the bar for what readers should expect in this space. Reinforcing that, partner support lets the tool recognize content from OpenAI, NVIDIA and Kakao systems, not only Google's — practical value rather than just headline appeal. The broader signal worth registering is straightforward: a watermark check answers whether SynthID is present, not whether the content is accurate, so it supports but does not replace verification. On the other side of the ledger, one constraint is real rather than a marketing footnote: content without a SynthID-compatible watermark, including local open-weight model output, cannot be flagged. It should factor into any serious decision. Layered on top of that, a login requirement and a quota of roughly 10 checks per day limit heavy or anonymous use — which narrows the set of teams for whom this is an obvious yes.

For journalists, educators, fact-checkers and anyone who regularly vets images, video or audio shared online, this is a serious evaluation candidate, not just a curiosity to bookmark. For everyone else, the safer posture is to monitor coverage and revisit once the use cases that matter to your team are demonstrated in the wild.