Google opens SynthID AI-content detector to users worldwide
In short
Google has opened its SynthID Detector to users worldwide, letting them check whether images, video or audio were generated by AI. The tool recognises watermarks from Google models and from partners including OpenAI, Nvidia and Kakao, with Apple reported to join later, and it is already built into the Gemini app and Chrome, where Google says about 1 million verification requests are made each day. Google says the results are not 100% accurate.
Read the full story 2 min read
Google has opened its SynthID Detector to users around the world, letting them check whether an image, video or audio file was generated by artificial intelligence. Jiemian, IT Home and cnBeta all reported the global opening, saying the tool now covers detection of content produced by Google's own AI tools as well as those of partners. [ 1 , 2 , 3 ]
SynthID was launched in 2023 as a digital watermarking technology, IT Home and cnBeta reported. Rather than a visible mark, the watermark is embedded during generation so that it is hard for people to see or hear, but can still be identified later. Google uses it in models including Nano Banana for images, Veo for video and Lyria for audio, and in platforms with content-generation functions such as Gemini, Flow, ProducerAI and Vids, according to cnBeta. Verification has also been integrated into the Gemini app and Google Chrome, and Google says users now make about 1 million verification requests a day. [ 2 , 3 ]
The tool has moved beyond Google's own ecosystem, the reports said, with OpenAI, Nvidia and South Korean internet company Kakao already supporting it, and Apple expected to join in the future. Google said the upgraded service can now recognise watermarks from partner systems, not just its own models, according to cnBeta, which added that OpenAI still keeps its own content-detection platform as a separate verification channel. [ 1 , 2 , 3 ]
cnBeta listed the file types the detector accepts: JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, TIF and GIF for images; MP4, MOV and WEBM for video; and WAV, MP3, OGG, FLAC, AAC and M4A for audio. [ 3 ]
Google also acknowledged limits. IT Home and cnBeta reported that the results are not 100% accurate, and cnBeta said watermark-based detection can only identify content produced with SynthID or other compatible standards, leaving AI works made outside those systems potentially undetected. Different companies still use different marking and verification systems, and no unified industry standard has emerged, it said. Microsoft and Meta are pursuing their own watermarking and verification schemes, cnBeta added, and industry participants generally consider these tools to have limitations. [ 2 , 3 ]
The documents differ on when the announcement was made. Jiemian dates it to October 7, while IT Home, publishing on the evening of October 7, says Google announced it “yesterday”. cnBeta adds that the tool was not entirely new, having been opened in a test version to selected journalists, media organisations and researchers during Google's I/O developer conference last year before this public release. [ 1 , 2 , 3 ]
Why it matters
A detection tool from a major platform, offered free to global users, gives ordinary readers another way to check whether media they see was AI-generated as synthetic images, video and audio spread. Its usefulness is limited by the fact that it only reads watermarking standards it and its partners use, and the documents say the industry has not settled on a single standard. Rival efforts by Microsoft and Meta mean users still face several, partly separate verification systems.
Key facts
- Google opened its SynthID Detector to global users, allowing them to check whether images, video or audio were AI-generated. [ 1 , 2 , 3 ]
- SynthID was introduced in 2023 as an invisible watermark embedded by Google models including Nano Banana, Veo and Lyria. [ 2 , 3 ]
- OpenAI, Nvidia and Kakao already support the tool, and Apple is expected to join in the future, according to the reports. [ 1 , 2 , 3 ]
- SynthID verification is integrated into the Gemini app and Google Chrome, where Google says about 1 million verification requests are made daily. [ 2 , 3 ]
- Google says the detection results are not 100% accurate and the tool can only identify content produced with compatible watermark standards. [ 2 , 3 ]
- The detector supports image formats including JPG, PNG, BMP, WEBP, AVIF, HEIC, TIFF and GIF, video formats MP4, MOV and WEBM, and audio formats WAV, MP3, OGG, FLAC, AAC and M4A. [ 3 ]
Confirmed by several sources
- Google has made its SynthID Detector available to users globally for checking whether images, video or audio were AI-generated. [ 1 , 2 , 3 ]
- The tool works with content generated by Google models and by partner systems, with OpenAI, Nvidia and Kakao already supporting it. [ 1 , 2 , 3 ]
- Apple is reported to be joining the supported partner list in the future. [ 1 , 2 , 3 ]
- SynthID dates from 2023 and is an invisible watermark rather than a visible mark on the content. [ 2 , 3 ]
- Detection has been built into the Gemini app and Google Chrome, and Google puts daily verification requests at about 1 million. [ 2 , 3 ]
- Google acknowledges the detection results are not fully accurate. [ 2 , 3 ]
Still unclear
- The exact date of Google's announcement. Jiemian dates the announcement to October 7, while IT Home, publishing on the evening of October 7, says Google announced it “yesterday”; the documents disagree.
- When Apple will join the supported partners, and on what terms. All three documents say only that Apple will join in the future, with cnBeta attributing it to reports (“据称”); no date or confirmation from Apple is given.
- How often the tool fails or what share of AI content it misses. Google is quoted as saying the results are not 100% accurate, but no figures or test results are provided.
- The earlier test release to journalists, media organisations and researchers at last year's Google I/O. This detail appears in only one document (cnBeta).
- Microsoft's and Meta's own watermarking and content-verification efforts. Mentioned only by cnBeta, without specifics on the systems or their progress.
What local media are saying
Timeline, local time
- Jiemian reports that Google has expanded the use of its AI-content detection tool SynthID, opening the SynthID Detector to global users. [ 1 ]
- IT Home reports the global opening and adds that SynthID verification is built into Gemini and Chrome and that daily verification requests reach about 1 million. [ 2 ]
- cnBeta publishes further detail on supported file formats, partner support, and the technical limits of watermark-based detection. [ 3 ]