Google Opens SynthID Detector Globally — But It Only Sees Watermarks It Knows

Google has made its SynthID Detector publicly available worldwide, letting anyone with a Google, Apple or OpenAI account upload images, video or audio and check within seconds whether it carries an AI watermark.

Google says SynthID has now watermarked more than 180 billion images and videos and 240,000 years of audio content.

What it can and cannot tell you

The tool detects invisible watermarks embedded by Google and participating companies — currently OpenAI, NVIDIA and Kakao, with Apple expected to join.

That is the limitation the announcement does not foreground, and it is the one users most need to understand.

A result showing no watermark does not mean content is authentic. It may have been produced by a provider outside the scheme, generated with an open-weight model run locally, or had its watermark removed. Meta, xAI, Stability and Midjourney are not currently listed among participants, and the open-weight ecosystem — which African developers increasingly use — sits outside it entirely.

SynthID Detector answers one question reliably: was this made by a participating provider? It cannot answer the question most users will bring to it: is this real?

The Nigerian election pitch

Google’s West Africa Communications and Public Affairs Manager, Taiwo Kola-Ogunlade, framed the release around the coming vote.

“With Nigeria’s next election cycle approaching, the tool could also provide an additional means for journalists, media organisations and members of the public to scrutinise suspicious AI-generated images, videos and audio circulated online,” he said.

The use case is real. Nigeria’s Defence Headquarters disowned an AI-generated video of its Chief of Defence Staff making partisan political statements in August, and Nigeria votes in 2027.

But the detector’s coverage is the question. A fabricated video of a Nigerian official is unlikely to have been produced on a watermarking platform by someone intending to deceive — and a tool that returns nothing on unwatermarked content gives a journalist no finding either way.

It is also English-only at launch, in a country where political content frequently circulates first in Hausa, Yoruba, Igbo and Pidgin.

How it developed

Google introduced SynthID in 2023 to watermark AI-generated content with imperceptible digital markers. An early version of the detector was released last year for media professionals before this week’s expansion to the general public.

The company describes the broader rollout as part of its effort to help people navigate an increasingly AI-driven online environment with greater confidence.

That is a reasonable description of what it does. It works best as one input among several — confirming provenance where a watermark exists, rather than establishing authenticity where one does not.