Google has dropped the waitlist for its SynthID Detector, so anyone can now upload an image, video, or audio file and see whether it carries Google’s invisible AI watermark. Search Engine Journal’s Matt G. Southern reported the change on October 7, citing a blog post by Google DeepMind’s Pushmeet Kohli, who leads science and strategic initiatives. The tool works globally, in English only.
An early version launched last year. Access then was limited to journalists, media professionals, and researchers who joined a waitlist. The public release moves the check from a gated press tool to something any marketing or social team can run before a file ships.
SynthID is the watermark Google embeds in media produced by its generative AI products. The detector lives at synthid.com, where users upload a file. Google’s post says the tool can identify files that Google or a partner produced with AI. The named partners are OpenAI, NVIDIA, and Kakao. Southern also reported that Apple is the one additional partner Google has flagged as upcoming, and no launch date was given.
The site is blunt about scope. In the words Southern quoted from it: “This is not a general AI detector. We can only detect media from the above companies, who have adopted SynthID technology.” The site also asks users to submit the highest-quality copy available, since that yields the most accurate reading.
Results carry their own caveats. In the sample video in Google’s post, the result panel reads “SynthID was detected” and “This media was made or edited with Google AI,” and adds that the media “may have been edited further since then.” The site’s FAQ says “no signal is completely robust” and recommends “collecting multiple points of data before making a decision.”
Southern flagged two limits that follow from that wording. A photo retouched with Google’s AI and an image generated from nothing by it could both return the same label, because the result says “made or edited.” And a file with no watermark found may still be AI output from a company that does not use SynthID.
For content and social teams, that asymmetry is the practical point. A positive result is useful evidence that a supported tool touched the file. A negative result proves only that no supported watermark was found. It does not show that a human made the asset, and it should not be written into a disclosure policy as if it did.
Sensible checks before publishing look like this:
- Run supplied or sourced images, clips, and audio through the detector, using the original, highest-quality copy rather than a compressed re-export.
- Record the result next to the asset’s provenance notes, including the source and the license terms.
- Treat the result as one data point, as Google’s own FAQ advises, alongside contributor statements and contract terms.
- Do not label a file as human-made on the strength of a negative result.
The release also sets up a contrast with text. Our October 6 report on OpenAI’s EU text watermark covered a detector limited to approved researchers. Google’s tool is open to the public, but per Southern’s reporting it covers images, video, and audio.
Neither Southern’s report nor Google’s post says how, or whether, SynthID results feed into any search or ranking system, and this article makes no such claim. The change is about asset hygiene, not visibility.
Teams that buy stock imagery, commission freelancers, or accept user-submitted media should add a SynthID check to the pre-publication checklist now. They should write the policy around what the tool can actually say.
Reported by Matt G. Southern of Search Engine Journal on October 7, 2026, based on a Google DeepMind blog post by Pushmeet Kohli.
