Google opened SynthID Detector to everyone on October 7, 2026. The website tells you whether an image, a video or an audio file carries the invisible watermark added by Google's AI models and by those of OpenAI, NVIDIA and Kakao. It is useful for anyone who publishes images on a business website and wants to know where they came from.
What we know
- What it is and when it opened: Google DeepMind announced on October 7, 2026 that SynthID Detector is now open to everyone. The post is signed by Pushmeet Kohli, a vice president at Google DeepMind. The address is synthid.com.
- Before: last year Google introduced an early version aimed at media professionals, according to the same post.
- Where and in what language: worldwide and in English, according to the post.
- What it checks: images, video and audio. For video and audio it points to the segments that carry the watermark, according to the tool's FAQ. It does not check text.
- Whose content it detects: content from Google and from the companies that have adopted SynthID: OpenAI, NVIDIA and Kakao. Apple is coming soon, according to the post, which gives no date.
- What it is not: the page itself says so: "This is not a general AI detector."
- Google's numbers: since 2023 it has watermarked more than 180 billion images and videos and 240,000 years of audio. The verification built into Search, the Gemini app and Chrome handles more than 1 million requests a day. These are the company's own figures.
- File formats: JPG, PNG, BMP, WEBP, AVIF, HEIC, TIFF and GIF for images; MP4, MOV and WEBM for video; WAV, MP3, OGG, FLAC, AAC and M4A for audio, according to the page.
- Getting in: you have to sign in with a Google, OpenAI or Apple account, according to Engadget and Android Authority. The tool's privacy notice mentions daily usage quotas and does not give the number.
- What happens to the file: according to that notice, it is deleted as soon as the result is returned. A digital signature of the file, tied to the user, is kept for 24 hours to enforce quotas and prevent abuse.
What changes and what doesn't
Who can check has changed. It is no longer a tool for media professionals: anyone with an account can upload a file and see the result.
The reach has changed. A single site reads the watermark of several companies, not only Google's.
The basic limit has not changed. The tool only looks for the SynthID watermark. If the image came from a model made by a company that does not use it, or from before that company adopted it, the result will be negative even if the image is AI-made. The same can happen when a file has been heavily compressed or edited. The FAQ says it plainly: a negative result does not rule out AI.
It is not infallible the other way either. In very rare cases it can trigger on content with no watermark, according to the same page, which recommends gathering more than one piece of evidence before deciding.
It is not built for checking a whole media library. Its terms forbid automated submissions, bulk queries and getting around the quotas.
It is not the only one. OpenAI runs its own verification page, which looks for C2PA metadata and the SynthID watermark in images and audio made with ChatGPT, its API or Codex.
How to tell if it affects you
It affects you if your website carries images you did not make yourself (from a supplier, a designer, a stock library or a customer), or if you generate your own with Gemini or ChatGPT. Seven steps:
- Pick what to check. There is a daily quota and no automation. Start with the images that ask for trust: product photos, team photos, premises, finished work and anything shown next to reviews or testimonials.
- Upload the original. The page asks for the highest-quality file you have. A screenshot, or a copy forwarded through a messaging app, has been recompressed and may have lost the signal.
- Read a positive carefully. "Watermark detected" means the file was probably generated or edited with a model that uses SynthID. It does not separate a fully generated image from a real photo retouched with AI.
- "Watermark not detected" does not mean "real photo". It only means no readable SynthID watermark was found. If the image matters, ask the supplier for the original file and ask in writing how it was made.
- Don't accuse anyone on one test. If the result does not match what you were told, ask before you complain. The page admits false positives, even if very rare.
- If you generate your images with Gemini or ChatGPT, assume anyone can check. Do not present an AI-made image of a product, a team or a customer as a real photo.
- Text is out of scope. The site does not analyze articles or product descriptions: only images, video and audio.
Related: ChatGPT to Add an Invisible Text Watermark in the EU
Sources
- Google, "Google expands SynthID Detector for AI content", Pushmeet Kohli, October 7, 2026
- Google, SynthID Detector: tool, FAQ, terms and privacy notice
- OpenAI, content verification page
- Engadget, "Google's AI detection website is now available", October 7, 2026
- Android Authority, "Google's cross-model AI detection tool is finally available for all", October 7, 2026
- TechCrunch, "Google's new SynthID website can identify AI-generated media", October 7, 2026
Updates: this note will be extended if Google publishes the daily quota, adds languages or confirms a date for Apple.