How to check a file for SynthID: official steps
Only the companies that use SynthID can read it, so the checks below all run on their own services. Google's blog says its SynthID Detector at synthid.com is open to everyone, in English, and covers media made with Google's AI tools and with partner tools, including OpenAI, NVIDIA, Kakao and Apple.
- Use the original file or the best copy you have. A screenshot, heavy crop or strong compression can weaken the watermark.
- Go to synthid.com, upload the image, video or audio file and read the result. For video and audio it marks the segments that carry a watermark.
- For a second route, open the Gemini app, add the file and ask “Is this AI-generated?”. Gemini recognizes only content made with Google's own AI tools, takes one file of up to 100 MB, and needs videos shorter than 90 seconds; for a screenshot, crop tight around the image.
- For an image or audio file you think came from OpenAI, also try OpenAI's verifier at openai.com/verify, which looks for SynthID and OpenAI's Content Credentials.
- Write down which service you used and what it said, and keep the file you checked: the result applies to that copy only.
Source: Google: SynthID Detector for everyone (updated October 29, 2026) (opens in a new tab)
What Image Evidence can read instead
Our checker on this page does not read SynthID and does not send your file to Google or OpenAI. It reads other provenance signals inside your browser tab, without uploading the image, so you can see what the file itself declares before or after an official SynthID check.
- C2PA Content Credentials: whether a manifest is present, whether it validates and which organization claims to have signed it.
- IPTC and XMP digital source type: whether the metadata declares AI generation or an AI edit.
- PNG text blocks with Stable Diffusion style generation parameters.
- Not read: SynthID, Meta's invisible watermark or any other in-pixel watermark, and the look of the picture itself.
What SynthID is
SynthID is an invisible watermark that Google's AI tools embed in generated media, distinct from ordinary file metadata. Google says it has watermarked more than 180 billion images and videos and 240,000 years of audio, and partner companies including OpenAI, NVIDIA, Kakao and Apple have adopted it. Because the signal lives in the content itself, an EXIF viewer cannot see it and a visual impression cannot confirm it. Image Evidence has no SynthID detector integration.
Source: Google: SynthID Detector for everyone (October 2026) (opens in a new tab)
About Google's SynthID Detector
Google opened its SynthID Detector at synthid.com to everyone in October 2026. It reports whether it found a SynthID watermark and covers media from Google and from the companies that have adopted SynthID. The site has its own terms, limits how many checks you can run each day, and says it deletes uploaded files once it returns a result. Image Evidence does not send files to it on your behalf.
Other official routes
The Gemini app runs a similar check when you upload a file and ask “Is this made with AI?”, but Google's help page says it currently recognizes SynthID only in content made with Google's own AI tools. It takes one file of up to 100 MB at a time, and videos must be shorter than 90 seconds. For images and audio made with OpenAI tools, OpenAI runs its own verifier at openai.com/verify, which looks for SynthID and for OpenAI's Content Credentials. Each service has its own terms and data handling.
Source: Gemini Apps Help: verify AI-generated media (opens in a new tab)
About Google's SynthID Detector
Google opened its SynthID Detector at synthid.com to everyone in October 2026. It reports whether it found a SynthID watermark and covers media from Google and from the companies that have adopted SynthID. The site has its own terms, limits how many checks you can run each day, and says it deletes uploaded files once it returns a result. Image Evidence does not send files to it on your behalf.
Interpret a missing signal
A watermark check can only answer questions within the detector's supported coverage. Google's detector says a “not detected” result does not rule out AI: the file may come from a model without SynthID, from before a company adopted it, or from a copy edited heavily enough to degrade the watermark. Do not treat an absent watermark as proof of a real scene, and do not equate a found watermark with a truthful caption.
SynthID versus C2PA
SynthID and C2PA answer different questions. C2PA Content Credentials are signed metadata that can describe capture, generation and edits, but they can be stripped when a file is re-saved or screenshotted. A SynthID watermark is embedded in the content and is designed to survive some common changes, but it only reveals that a SynthID-using tool was involved. A file can carry both signals, one of them or neither.
Keep verification responsible
Ask the owner for an original file when appropriate and preserve the claimed source in your own notes. Use only material you are entitled to process, especially when images contain private information. Image Evidence does not identify faces, locate people, train on user content or remove watermarks, and it does not help anyone bypass detection. For an important dispute, consult multiple sources and record unresolved questions instead of presenting a provisional result as a certificate.
How do I check if an image has a SynthID watermark?
Upload the original file to Google's SynthID Detector at synthid.com, which looks for watermarks from Google and from partners such as OpenAI, NVIDIA, Kakao and Apple. The Gemini app offers a similar check for content made with Google's own tools. Use the highest-quality copy you have, and remember that a result applies to that file, not to every copy of the image.
Can SynthID Detector check ChatGPT images?
Yes, if the watermark survived. OpenAI says images generated with ChatGPT, Codex and its API carry a SynthID watermark as well as Content Credentials, and Google lists OpenAI among the companies its detector covers. Screenshots, heavy crops and strong compression can weaken the watermark. OpenAI's own verifier at openai.com/verify can check the same file, and our ChatGPT image guide covers the Content Credentials side.
Can SynthID be detected after an image is edited?
Google designs SynthID to remain detectable after some common changes, such as filters or compression, so a watermark may still be found after modest edits. Heavy edits can weaken it, and no watermark survives every change. A negative result on an edited copy is therefore less informative than one on an original.
Does a SynthID result mean an image is fake?
It means a SynthID-watermarking AI tool was likely used somewhere in the file's history. The image may be a fully generated illustration, a lightly edited photo or openly disclosed creative work. Whether it is misleading depends on how it is presented, which a watermark cannot judge. Google also says SynthID can, in very rare cases, trigger on content that was never watermarked, so look for corroborating evidence when the result matters.