A convincing video or an image that has already been shared several times can be difficult to trace from visual details alone. On October 7, Google announced public access to SynthID Detector, with a globally available English-language entry point for checking supported AI watermarks in images, video and audio.
The main change is who can access the detector. SynthID itself is not new: Google introduced it in 2023, and an earlier detector was initially offered to selected media professionals. Creators, editors and ordinary readers now have a more direct way to check for these signals.
It checks a watermark, not whether something looks like AI
SynthID adds a marker during generation that is designed to be imperceptible to people; detection looks for that marker. Google’s announcement describes checking content from Google and participating partners, including OpenAI, NVIDIA and Kakao. Apple support is described as coming soon.
This does not mean every AI model or every version is covered. A company appearing on the partner list does not establish that all content it has generated in the past carries a detectable watermark.
The result that is easiest to misread: no watermark found
Not finding a particular watermark means only that this check did not find it. Content could come from a model that does not use the technology, or subsequent processing and other factors could affect the result. It does not prove that a human created the content, or that the event shown actually happened.
A detected AI watermark does not establish malicious intent, either. Advertising assets, product demonstrations and creative films can legitimately use AI. Whether content is misleading still depends on the publisher’s explanation, its context and its purpose.
How content teams can use it
AI Islands recommends treating the detector as one clue in a source-checking workflow. When you receive material of uncertain origin, preserve the original file and earliest link, then try detection. Check the publishing account, original source and independent corroboration as well. For important events or statements attributed to people, a screenshot of a detection result should not be the sole basis for a conclusion.
For example, suppose you want to share a video presented as a live demonstration of a new product. Even if no watermark is detected, check whether the brand published the same demonstration and whether the recording has a credible source. If a watermark is found, check whether it is a clearly labeled concept video. Establishing provenance and establishing truth are different tasks.
Access and limits
The official public entry point is synthid.com. The announcement specifies an English-language release; it does not establish that a Chinese interface is available. Check the tool’s current instructions for upload requirements and account conditions.
If you create AI images or video, explain whether generative AI was used and which parts were edited. The AI tool directory can help you find creation tools. Watermarks help trace provenance, while clear creation notes help audiences understand the content. Neither replaces fact-checking.