Google opens SynthID Detector for AI-made images, video and audio to all users
In short
Google on October 7 (US time) opened SynthID Detector to all users. The website checks uploaded images, video and audio files for SynthID, the watermark Google DeepMind puts into AI-generated content, and is available only in English. Besides Google's own AI, it covers content from OpenAI, NVIDIA and Kakao AI that use SynthID, and support for Apple is due soon. Until now only a limited group of users could access it.
Read the full story 2 min read
Google on October 7, US time, opened SynthID Detector to all users, Japanese tech outlets Mynavi News, ITmedia and Gigazine reported. The website checks image, video and audio files for SynthID, an invisible watermark added to AI-generated content, and is available only in English. Users go to synthid.com and upload a file. According to Mynavi News, the tool then shows whether a watermark was found: for images it reports whether one is present, and for video and audio it shows which segments carry it. [ 2 , 3 , 4 ]
Besides content from Google's own AI, the detector checks content from OpenAI, NVIDIA and Kakao AI that use SynthID, and support for Apple is due soon, Mynavi News and ITmedia reported. ITmedia said watermark checks were already possible in the Gemini app but only for content made with Google's AI. Mynavi News reported that Google has built SynthID checks into the Gemini app, Google Search and Chrome, which now handle more than one million requests a day. [ 2 , 4 ]
Gigazine tested the service and reported that users must sign in with a Google, Apple or ChatGPT account. An illustration made with Gemini was flagged as carrying SynthID and was correctly identified as made with Google's AI. A version made with ChatGPT was reported as made or edited with OpenAI. When that image was cropped to show only the cat and its brightness was edited, SynthID was still found, but the result no longer named OpenAI. A panel from a comic that was not made with AI showed no watermark. [ 3 ]
All three outlets noted the tool's limits. Content from AI that does not use SynthID is not detected, so a negative result does not show that a person made the content. ITmedia said Google warns on the site that it is not a general-purpose AI detection tool, and that the terms of use set a daily limit on checks without stating the number. Mynavi News said heavy editing may make a watermark undetectable. The outlets disagree on text: Mynavi News said the Detector does not support SynthID's text watermark, while Gigazine described the portal as accepting text. [ 2 , 3 , 4 ]
SynthID was developed by Google DeepMind and announced in 2023. Google says it has since been applied to more than 180 billion images and videos and to audio totalling 240,000 years. According to Mynavi News, the image watermark is designed to survive filters, colour changes and lossy compression such as JPEG. Gigazine said the watermark sits in the image pixels, so it survives when metadata is removed or the image is cropped. Google announced SynthID Detector in May 2025, when it was open only to journalists, researchers and other testers, some of whom joined through a waitlist. [ 2 , 3 , 4 ]
Why it matters
Users can now check content made with AI from several companies, not only Google's, which ITmedia said had been the limit of the earlier check in the Gemini app. All three outlets noted that the tool only finds SynthID watermarks, so content from AI that does not use SynthID is not detected and a negative result does not prove a person made it.
Key facts
- Google opened SynthID Detector to all users on October 7, US time. [ 2 , 3 , 4 ]
- The tool checks image, video and audio files for SynthID watermarks and is offered only in English. [ 2 , 3 , 4 ]
- It detects content from Google and from OpenAI, NVIDIA and Kakao AI that use SynthID; support for Apple is due soon. [ 2 , 4 ]
- SynthID Detector was announced in May 2025 and had been limited to journalists, researchers and other testers. [ 2 , 3 , 4 ]
- Google says it has applied SynthID to more than 180 billion images and videos and to audio totalling 240,000 years. [ 2 , 3 ]
- Mynavi News reported that SynthID checks in the Gemini app, Google Search and Chrome handle more than one million requests a day. [ 2 ]
- Content from AI that does not use SynthID is not detected, so a negative result does not prove a person made the content. [ 2 , 3 , 4 ]
Confirmed by several sources
- Google opened SynthID Detector to all users on October 7, US time. [ 2 , 3 , 4 ]
- Access had previously been limited to a smaller group of users after the tool was announced in May 2025. [ 2 , 3 , 4 ]
- The detector covers content generated by OpenAI, NVIDIA and Kakao AI that use SynthID, and support for Apple is due soon. [ 2 , 4 ]
- The service is available only in English. [ 2 , 3 , 4 ]
- Google says SynthID has been applied to more than 180 billion images and videos and to 240,000 years of audio. [ 2 , 3 ]
- Content from AI that does not use SynthID is not detected, and the absence of a watermark does not show that the content was made by a person. [ 2 , 3 , 4 ]
Still unclear
- Whether SynthID Detector checks text. Mynavi News says SynthID has a text watermark but the Detector does not support it, while Gigazine describes the portal as accepting text along with images, audio and video.
- How many checks a user may run each day. ITmedia reports that the terms of use set a daily limit but do not state the number.
- When Apple's AI content will be covered. Mynavi News and ITmedia say only that support for Apple is coming soon.