Google opens SynthID Detector to everyone, now checking watermarks from OpenAI, NVIDIA and Kakao, with Apple next

Why it matters

Key points

Google opened SynthID Detector, its watermark checking tool, to everyone on the evening of October 7 Beijing time (October 7, 14:00 UTC). It is available globally in English at synthid.com. Upload an image, video or audio file and it tells you whether the content carries a SynthID watermark, meaning it was generated or edited by AI from Google or one of its partners.

The biggest change is that it now works across companies. Besides Google's own Gemini, Imagen, Veo and Lyria, it recognizes watermarked content from OpenAI, NVIDIA and Kakao, with Apple support "coming soon." Last year the tool was limited to media professionals and only recognized Google's own watermark.

Google also shared scale figures: since launch, SynthID has watermarked more than 180 billion images and videos and 240,000 years of audio, and the verification features built into Search, the Gemini app and Chrome now handle more than 1 million requests a day.

Verifiable

Official post: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/synth-id-ai-content/ , by Pushmeet Kohli, VP of Science and Strategic Initiatives at Google DeepMind; the structured publish time corresponds to 22:00 on October 7 Beijing time.

According to same-day reports from Ars Technica and The Verge, you need to sign in with a Google, OpenAI or Apple account; each user gets roughly 10 checks a day; and the result only says whether SynthID was used to generate or edit the media, without the region highlighting of the earlier internal version. Unite.AI quotes the site's FAQ as saying it is "not a general AI detector" and only recognizes content from companies that have adopted SynthID.

Our take

What matters here is that SynthID is turning from a Google technology into a de facto industry watermark standard. OpenAI and NVIDIA letting Google's tool check their watermarks shows the leading companies have chosen shared infrastructure, not separate systems, for proving content is AI-made. Together with OpenAI's recently published approach to EU text provenance rules, content provenance is moving from voluntary pledges by individual companies toward a cross-company network.

Readers should understand the limits, though. Finding a watermark is strong evidence that a participating company's AI made the content; finding none proves nothing. Content from open-weight models or tools outside SynthID, or files that have been screenshotted or re-encoded, may not be detected. The roughly 10 daily checks and the sign-in requirement also show this is a verification entry point for ordinary users, not a tool for moderating content at scale.

Two things to watch next: when Apple actually joins, and whether major Chinese model makers adopt the same system. If only US companies take part, a "no watermark found" result says almost nothing about the large volume of AI content on the Chinese-language internet.