SynthID detector
Upload an image or an audio file to check it for a supported SynthID watermark. The check runs through the OpenAI Content Provenance API, and the result is reported on its own rather than merged with C2PA Content Credentials.
Upload a file to check itUpdated 7 August 2026What SynthID is
SynthID is a watermarking technique from Google DeepMind. It embeds a marker directly into an image's pixels or an audio file's waveform, below the threshold of human perception, rather than writing it into the file's metadata.
That distinction matters. Metadata is a layer wrapped around a file, and almost any re-save can drop it entirely. A SynthID marker is distributed through the content itself, so as long as the picture or the sound survives, the marker has a chance of surviving with it.
How this checker reads the signal
Your file is sent to the OpenAI Content Provenance API, which reports whether a supported SynthID watermark is present. This site does not train its own model and does not guess from style, texture, or brushwork.
That makes the result grounded and binary: either a supported watermark was read, or it was not. You will never see a score like "78% likely AI" here.
What weakens or destroys the watermark
SynthID is more durable than metadata, but it is not indestructible. The signal degrades as the content is altered.
- Images: screenshots, photographing a screen, heavy cropping, repeated compression or re-encoding, strong colour grading and filters
- Audio: resampling, speed changes, layered background noise, keeping only a very short excerpt
- Both: shrinking the file to a very small size, or passing it through several rounds of platform transcoding
- Separately, content from earlier models may never have carried a SynthID watermark at all
SynthID compared with C2PA
The two are complementary signals, not alternatives. C2PA Content Credentials are signed metadata attached to the file, carrying rich context such as the issuer, the model, and the creation time — but they are easily stripped in one step when a file is shared, converted, or edited.
SynthID lives inside the content, carries none of that detail, and is far more likely to survive compression and moderate editing. That is why this site reports the two results separately and lets you weigh the evidence yourself.
What this checker cannot do
This site detects supported OpenAI provenance signals through OpenAI's API. SynthID as a technology is also used in Google's own products, but that content falls outside what this checker covers.
No watermark found does not mean a human made it. The file may come from another company's AI tool, from an earlier model, or it may have lost the signal somewhere along the way.
FREQUENTLY ASKED
Is this SynthID checker free?
Yes. You can upload a file and check it without an account and without payment. One file at a time, up to 50 MiB, with audio no longer than 60 seconds once decoded.
Can it detect content made with Google Gemini?
No. Although SynthID is Google's technology, this site checks through OpenAI's API and covers supported OpenAI provenance signals only. Content from Gemini or other tools will come back as no supported signal detected.
Can SynthID still be detected after a screenshot?
Sometimes, but the success rate drops sharply. A screenshot re-encodes the image and usually rescales or crops it, all of which weaken the watermark. For a reliable answer, use the original file as you first received it.
What does a detected SynthID watermark tell me?
That the file very likely came from a supported OpenAI tool. It is solid evidence of origin, but it is not the file's full history, and it says nothing about whether the content is truthful or presented in context.
Which file formats are supported?
PNG, JPEG, and WebP for images; MP3, Opus, AAC, FLAC, WAV, and PCM for audio.