Google opened global access to SynthID Detector on Wednesday, October 7, allowing people to check images, videos and audio produced with compatible AI tools. The service is available in English and covers content from Google and partners including OpenAI, NVIDIA and Kakao. Apple will be added later, according to Google's announcement.
The company reports watermarking more than 180 billion images and videos, alongside the equivalent of 240,000 years of audio. Its verification features in Search, Gemini and Chrome collectively receive more than one million requests a day.
Public access follows testing with media professionals
Google first introduced the SynthID Detector portal in May 2025. At that stage, it offered access to early testers and invited journalists, researchers and media professionals to join a waiting list. The description of that version explained a process in which users uploaded material for digital watermark scanning. Results could identify areas within an image or audio segments containing a detected watermark. The portal brought checks for several types of content together in a shared interface designed to examine the provenance of generated media.
SynthID, developed by Google DeepMind, embeds watermarks directly into content generated or altered with artificial intelligence. These signals are designed to be imperceptible to people while remaining recognizable to detection systems. The technology overview covers images, audio, video and text. Watermarking takes place within products that implement it, in connection with content generation. A detector subsequently searches for that signal; identification depends on a compatible watermark being present in the material under examination. The system's stated purpose is transparency about the use of AI to produce or modify content.
Digital watermarking has technical limits
At SynthID's launch in August 2023, DeepMind described two deep learning models trained together: one to apply the watermark and another to identify it. The system was designed to preserve the signal after operations such as changing colors, applying filters or JPEG compression. The team nevertheless cautioned that the technology was not guaranteed to withstand extreme manipulation. Because it is embedded in pixels, the watermark can remain detectable when file metadata is removed. This distinguishes the signal inside the image from descriptive information stored alongside it.
Google has also documented specific limits involving edited photographs. In February 2025, the company announced SynthID watermarking for images modified with Reimagine in Magic Editor, within Google Photos. Its announcement at the time explained that a very small intervention, such as changing the color of a flower in the background, might not produce a detectable watermark. The system was therefore extending from fully AI-generated images to photographs edited using generative tools. The size and nature of an alteration remain relevant to whether the automated intervention can be identified.
Gemini's documentation also distinguishes verification inside the app from other implementations. In the description available at the time of writing, SynthID verification in Gemini recognizes content made with Google AI tools. A result reporting no detected watermark does not rule out production using other AI systems. The page also describes inconclusive results, including images with insufficient detail or very small edits. Conditions documented for the Gemini app cannot automatically be applied to the new public portal or to every tool using the same underlying technology.
Search also provides context about where images appear
In Google Search, image provenance checks sit within a broader set of contextual information. About this image, also described in connection with its integration into Lens and Circle to Search, can show how other pages use and describe an image, including news coverage and fact-checking reports. It can display available metadata and information about SynthID watermarks. Google notes that metadata can be added or removed by the person publishing an image. This information describes the file's origins and use, without replacing examination of the claims attached to it.
Another component of verification is the Content Credentials standard developed by the Coalition for Content Provenance and Authenticity, or C2PA. The standard's explainer describes records cryptographically associated with an asset, which can document its origins, editing and use of artificial intelligence. These records aim to make unauthorized changes to provenance information identifiable. They can document tools involved in producing a file and the relationship between the finished asset and its constituent elements. Participation in the ecosystem is voluntary, and the standard seeks interoperability between applications and devices.
Google outlined its collaboration with C2PA in a September 2024 announcement. The company contributed to version 2.1 of the standard and described bringing provenance information into its products. For Search, the announcement concerned displaying C2PA information about AI image creation or editing through About this image. Google explained that protecting a piece of content's history requires validation rules and verification of the source signing the data. The initiative was presented alongside the expansion of SynthID as part of the company's tools for content transparency.
C2PA states that validating these records does not establish whether what an image depicts is true. The standard checks the association between provenance information and the file, its integrity, and trust in the signer. Even material with a verifiable history can require further examination of the place, date and circumstances in which it is presented. Detecting AI involvement and confirming the facts communicated through an image therefore remain separate operations, addressing different questions about the same piece of content.
