On October 7 Google opened SynthID Detector to everyone. The tool used to be for journalists only; now anyone can upload a photo, video or audio file to synthid.com and find out whether it carries the invisible watermark used by Google's AI and its partners. Here's how to check a file, what the detector sees, where it's useless and what it changes for anyone making creatives with AI.
What SynthID is, in plain words
SynthID is a Google DeepMind technology that since 2023 has embedded an invisible watermark into generated content. It isn't a metadata tag that disappears on re-save but a signal inside the image pixels or the audio itself. People can't see or hear it; the detector can find it.
Google says it has marked over 180 billion images and videos and about 240,000 years of audio. The mark comes from Google models such as Gemini, Nano Banana, Veo and Lyria, and now from partners too.
What SynthID Detector checks
- Images: 11 formats, including JPG, PNG, WEBP, HEIC and AVIF.
- Video: MP4, MOV and WEBM. For video and audio the detector shows which segment carries the mark.
- Audio: WAV, MP3, OGG, FLAC, AAC and M4A.
- Whose AI: Google, OpenAI, NVIDIA and Kakao. Google says Apple Intelligence support is coming "soon".
How to check a photo or video for AI with SynthID Detector
- Open synthid.com.
- Sign in with a Google, OpenAI or Apple account; checks don't work without signing in.
- Upload the file. Use the original at the highest quality: heavy compression and editing weaken the mark.
- Read the result: "watermark detected" or "not detected". For video and audio the marked segments are highlighted.
The service is free but English-only for now, and there's a daily check limit. Per the terms, uploaded files are deleted after the check. The same check is built into the Gemini app, Google Search and Chrome, which Google says handle over 1 million requests a day combined.
What SynthID Detector can't do
This matters more than it seems. Google itself says it is not a general AI detector:
- Text isn't checked in the portal. SynthID can mark Gemini text, but synthid.com only accepts images, video and audio. Claims that "you can now check any text" are wrong.
- It doesn't see other AI. If a generator doesn't apply SynthID, the detector says "not detected" even for a 100% AI image.
- "Not detected" doesn't mean "made by a human". No mark proves nothing.
- It can't tell "created" from "edited". A photo where AI only changed the background will also show the mark.
- The mark can be weakened. Google warns the signal is harder to read after heavy editing, and Engadget notes that in some cases it can be removed. Claims that the system is "impossible to fool" are overstated.
What it means for media buyers
If you generate creatives in Nano Banana, Veo or ChatGPT, the file almost certainly carries an invisible mark. Now anyone can check it: a moderator, a competitor or a user who doubts a "real testimonial".
- Social platforms already label AI content. Since 2024 TikTok has automatically labeled AI content using the C2PA standard, and Meta labels AI images on Facebook and Instagram. Neither has officially announced using SynthID, but the direction is clear: AI content is becoming visible.
- Don't pass AI off as real. A generated "customer video" or "before/after photo" without disclosure is a direct risk of ad rejection and bans under platform rules on misleading content.
- Check other people's creatives. Before taking a creative from a spy tool or a freelancer, run it through the detector so you know whether it's AI or real footage and don't pay for "original UGC" that was actually generated.
- Disclose AI yourself. If a platform asks you to flag AI content, do it. That's cheaper than losing the account.
If you work with short videos, see our guide TikTok UBT for beginners.
FAQ
How can I tell if a photo was made with AI?
Upload it to synthid.com: if it was made with Gemini, Nano Banana, ChatGPT or another supported model, the detector finds the mark. No mark isn't proof: the image may come from AI without SynthID.
Is SynthID Detector free?
Yes, but you need to sign in with a Google, OpenAI or Apple account, and there's a daily check limit.
Can I check text for AI with SynthID?
Not in the portal: it only accepts images, video and audio.
Does SynthID Detector work in other languages?
At launch the interface is English-only, but you can check files with any content.
Sources: Google blog, synthid.com, Engadget, MacRumors, PPC Land.