Google Opens SynthID Detector Worldwide: Check Any Photo for AI Watermarks

What Happened

Pushmeet Kohli, Google DeepMind's VP of Science and Strategic Initiatives, announced on the official Google blog (The Keyword) on October 7 (ET) that the SynthID Detector is now open to everyone worldwide (in English) at synthid.com — upload an image, video, or audio file to check whether it carries an invisible AI watermark. Supported formats: a dozen-plus image formats (JPG / PNG / HEIC / GIF and more), video in MP4 / MOV / WEBM, audio in WAV / MP3 / FLAC and more. Coverage: Google's own models (Gemini, Veo, Lyria, Nano Banana, plus generation-capable tools like Gemini, Flow, ProducerAI, and Vids) plus partners that adopted the SynthID standard — OpenAI, NVIDIA, and Kakao; Apple is "coming soon" per Google. Last year (Google I/O 2025) the tool was limited to a small group of journalists, media professionals, and researchers; now it works straight from the browser — though per webpronews, you need to sign in with a Google / Apple / OpenAI account first; anonymous checks are not supported.

Since SynthID launched in 2023, Google says it has watermarked more than 180 billion images and videos plus 240,000 years of audio — the official May 19 figures were 100 billion images/videos and 60,000 years of audio, so the volume nearly doubled in under five months. Detection was already built into Search, the Gemini app, and Chrome: the Gemini app's built-in verification had 50 million uses globally as of May, and the three entry points now handle over 1 million verification requests a day combined. The Detector site's FAQ states: "This is not a general AI detector." It only recognizes content from companies that adopted SynthID; media outside that coverage is not identified.

Key Facts

1. No watermark found does not mean the content is real: The SynthID Detector can only find content carrying a SynthID watermark. No watermark found means "no recognized watermark found" — it does not prove the photo was taken by a human. AI-generated but unwatermarked, watermarked to a rival standard, or watermarked but damaged by heavy cropping or compression — the Detector cannot identify any of them. This is stated in the official FAQ.
2. Where the "1 million requests a day" comes from: 180 billion watermarks, 240,000 years of audio, 1 million verifications a day — all Google's official figures, not independently audited. The "1 million a day" comes mostly from built-in verification in Search / the Gemini app / Chrome, not uploads to synthid.com; the site only fully opened today.
3. Watermark standards are fragmented: OpenAI, NVIDIA, and Kakao use SynthID, Apple is coming soon, while Microsoft and Meta each run their own watermarking standards: TechCrunch notes these detection tools "often fail to identify content created by their makers' models." Content one checker flags as AI-generated may not be flagged by another.
4. Two limits: sign-in required, English-only for now: (1) You need to sign in to upload (per webpronews, a Google / Apple / OpenAI account); anonymous checks are not supported; (2) it is English-only for now, with other languages coming later. The rollout: I/O early access last year, built into Search / Gemini / Chrome in May, fully public today.

Context

The backstory: SynthID launched in 2023 alongside Google's AI generation tools as an invisible-at-creation watermarking scheme; Google I/O 2025 gave journalists and researchers an early Detector build; on May 19, 2026, Google said the Gemini app's built-in verification had 50 million uses and expanded it to Search, then Chrome. Separately, yesterday (October 6) this site covered the textGrain watermarking scheme.