Google has launched a public SynthID AI detector, a website at synthid.com that lets anyone check whether an image, video, or audio clip was generated or edited with AI. Until this week, the detection side of SynthID was limited to journalists, media professionals, and researchers who requested access starting at Google I/O in 2025.
What the Public SynthID AI Detector Actually Checks
The new site accepts a wide range of file types. On images, it reads JPG, JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, and TIF, plus GIF. On video, it reads MP4, MOV, and WEBM. On audio, it reads WAV, MP3, OGG, FLAC, AAC, and M4A. A user uploads a file and the site reports whether it can detect a SynthID watermark embedded at the time the content was generated. Google first introduced SynthID in 2023 as a watermarking system, embedding an imperceptible signal into AI-generated output so it could be identified later. The public site is the detection half of that system, now opened beyond the small group that previously had access.
Which AI Tools Already Watermark Their Output With SynthID
Google's own generative tools, including Nano Banana, Veo, Lyria, Gemini, Flow, ProducerAI, and Vids, all apply SynthID watermarking to what they produce. SynthID verification is also built directly into the Gemini app and into Google Chrome, so users can check content without leaving those products. The standard extends past Google: OpenAI, Nvidia, and Kakao also support SynthID, and OpenAI runs its own separate verification site at openai.com/research/verify. Apple is reportedly adding support as well, according to TechCrunch's October 7 report.
How Much the Tool Is Actually Used
Google says people submit about 1 million verification requests through SynthID every day, a volume that points to real demand for a way to check AI content outside of specialized newsroom or research settings. That number also reflects how much synthetic media is now circulating widely enough that ordinary users, not just professional fact-checkers, want to verify it themselves.
Why Detection Tools Like SynthID Are Not a Complete Answer
Microsoft and Meta operate their own watermarking and verification systems, and none of these tools, including SynthID, catch everything. A Reuters report cited by TechCrunch found that Meta's own AI image detector failed to identify some of Meta's own AI-generated images once they had been cropped. That gap illustrates the core limitation of watermark-based detection: it depends on the watermark surviving whatever editing, compression, or cropping happens to a file after it is created, and no current system guarantees that.
Why This Matters for AI Governance and Procurement
Organizations adopting AI content-verification tools, whether for newsroom fact-checking, trust-and-safety programs, or internal compliance, should treat SynthID and its competitors as one signal among several, not a definitive answer. The Reuters finding on cropped images is a direct warning for any governance program that plans to rely on a single watermark check before publishing, approving, or flagging content. Engineering and policy teams procuring these tools need processes that assume detection will sometimes fail silently, with human review and multiple verification methods layered on top, rather than a single pass-fail gate tied to one vendor's watermark.
FAQ
What is Google’s SynthID and what does the new website do?
SynthID is Google’s system for watermarking AI-generated media. The new public site at synthid.com lets anyone upload an image, video, or audio file to check for that watermark, a feature previously limited to journalists and researchers.
Which file formats does the public SynthID tool support?
It supports image formats including JPG, PNG, WEBP, and HEIC, video formats MP4, MOV, and WEBM, and audio formats WAV, MP3, OGG, FLAC, AAC, and M4A.
Which AI companies use SynthID watermarking?
Google’s Nano Banana, Veo, Lyria, Gemini, Flow, ProducerAI, and Vids all apply it. OpenAI, Nvidia, and Kakao also support the standard, and Apple is reportedly adding support.
How many people use the SynthID verification tool?
Google says the tool handles about 1 million verification requests per day.
Can SynthID and similar tools be fooled or fail to detect AI content?
Yes. A Reuters report found Meta’s own AI image detector failed to identify some of Meta’s AI-generated images after they were cropped, showing watermark detection can fail once content is edited.
Is SynthID the only AI content-detection tool available?
No. Microsoft and Meta run their own watermarking and verification systems, and OpenAI operates a separate verification site for its own content.