Google has opened SynthID Detector to the public this week. The website lets anyone upload an image, a video clip or an audio file and see whether it carries SynthID, an invisible watermark that Google and some partners embed in AI-generated media. It is available worldwide in English, and you need to sign in to use it.
The launch is useful, but the tool answers a narrower question than its name suggests. It does not tell you whether something was made by AI in general. It tells you whether a specific watermark is present.
How SynthID works
SynthID is Google’s watermarking system, introduced in 2023. It hides a signal inside generated media that people cannot see or hear but software can read. Google’s own generators apply it automatically, including its image model Nano Banana, its video model Veo, its music model Lyria, and apps built on them such as Gemini and Flow. Google says it has watermarked more than 100 billion images and videos and 60,000 years of audio. These are company figures that nobody has independently verified.
A version of the detector has existed since Google I/O last year, but only a small group of journalists, media professionals and researchers could use it. Opening it to everyone turns what was a specialist tool into something a curious reader can try in a minute. Google also says people already make about a million verification requests a day through checks built into Gemini, Search and Chrome, so the website is a more direct route to a feature many people have already used without noticing.
What the site accepts and detects
The site handles the common formats. For images that includes JPG, PNG, WEBP, HEIC and GIF, for video MP4, MOV and WEBM, and for audio MP3, WAV, FLAC, AAC and M4A, among others.
It can detect media from Google’s tools and from companies that have adopted SynthID. That list currently includes OpenAI, Nvidia and Kakao. Apple is expected to join soon, which would let the site recognize images made with Apple’s AI features, such as Image Playground.
What it cannot do
The limits matter more than the launch itself. The site is not a general AI detector. If a file has no SynthID watermark, the result is negative, and a negative result is not proof that the file is real. It may have been made by a model that uses a different watermarking standard, or the watermark may have been removed or damaged. Google has acknowledged in the past that watermarks can be circumvented.
The tool also cannot tell content that was created entirely by AI from content that was only edited with AI. A photo that is genuine but retouched with an AI tool can come back positive, which can mislead anyone who reads a positive result as “this is fake.”
Google says on the site that it is working on more partnerships so that more kinds of generated media can be detected, and that industry-wide standards take time because images, audio and video have different technical requirements.
How to use it
Open the site, sign in with a Google account, and upload the file you want to check. A positive result means the file carries a SynthID watermark and was made or edited with a tool that uses it. A negative result means only that no watermark was found.
What this means for you
For everyday users, the tool settles one specific kind of doubt: whether a suspicious image, clip or voice recording came from an AI tool that marks its output. It works best as one signal among several, alongside the source of the file, the context in which it appeared and a plain check of whether anyone credible published it.
For people who make media with AI, watermarking is becoming a standard part of the toolchain, and a public detector makes that visible. Files you generate with participating tools may be identifiable even after you share them, so it is worth knowing which of your tools mark their output.
The same week, OpenAI announced that it will add an invisible watermark to ChatGPT and Codex text in the EU, which suggests that provenance labels are spreading from images and audio into writing. Expect more tools to adopt their own marks, and expect the mix of standards to stay messy for some time.
