What is SynthID?
Updated October 2026
SynthID is a watermark Google embeds into the pixels or waveform of AI-generated media, invisible to people and robust to cropping and re-encoding. Only Google can read it — but its public detector at synthid.com lets you check a file yourself, for content from companies that have adopted the technology.
Most ways of identifying AI content work on metadata attached alongside a file, which means a screenshot destroys them. SynthID takes a different approach: it alters the content itself, subtly enough that you cannot see or hear the difference, in a pattern that survives the handling that strips metadata.
That makes it the most durable signal in this field, and also the most limited — because reading it requires the key, and the key is Google's.
How it differs from Content Credentials
C2PA Content Credentials are a signed record attached to a file. SynthID is a change made to the content. The distinction decides everything about how each behaves.
A credential is readable by anyone with a verifier, carries rich information about what created the file and what was done to it, and is destroyed the moment the file is re-encoded. A watermark carries little more than “this was generated”, cannot be read without the detector, and survives cropping, compression, format conversion and screenshotting.
They are complements rather than competitors. Provenance tells you the detail; watermarking is what remains after the internet has had its way with the file.
Where you can actually check it
Google runs a detector at synthid.com. It accepts images, audio and video, and asks for the highest-quality copy of a file you have, since compression reduces detection accuracy.
The portal is explicit about its scope: “This is not a general AI detector. We can only detect media from the above companies, who have adopted SynthID technology.” Adoption extends beyond Google to several partners, and the site lists the current set.
Verification also runs inside the Gemini app, and Google has said it is extending it into Search and Chrome.
Why no third-party tool can read it
Detecting the watermark requires knowing the pattern it was embedded with. Publishing that would also tell anyone how to strip or forge it, so Google keeps it closed and offers detection as a service instead.
The practical consequence is worth stating plainly: any site offering to check SynthID is not doing so. This one cannot, and says as much. If you want a SynthID answer, Google's own portal is the only place to get one.
What a negative result does not mean
If the detector finds no watermark, that tells you the file was probably not produced by a participating company — not that it was made by a human. Most generative tools do not use SynthID at all, so an unwatermarked image is entirely unremarkable.
This mirrors the asymmetry across the whole subject. A positive signal — a watermark, a signed credential, a prompt embedded in a PNG — is strong evidence. The absence of any of them is weak evidence of nothing in particular.
Does SynthID work on text?
Google has published research on watermarking language-model output, and the approach biases word choice in a pattern a detector can recognise statistically. It is a genuine technique and an active research area.
Text watermarking is inherently more fragile than image watermarking, because there is far less room to hide a signal in a sentence than in a few million pixels, and ordinary editing disturbs it. Do not assume a passage of text can be checked the way an image can.
Try it on a file
Runs every check described above, in that order, and shows you which one produced the answer.
Drag a file here, or
JPEG, PNG, WebP, AVIF, PDF or .txt · up to 10 MB · you can also paste a screenshot
By uploading you agree to our Terms and Privacy Policy. Your file, IP address, approximate location and device details are stored.
Frequently asked questions
Can I check SynthID for free?
Yes, through Google's own detector at synthid.com, which accepts images, audio and video. It only recognises media from companies that have adopted SynthID, so it is not a general-purpose AI detector.
Does a screenshot remove SynthID?
Generally no, and this is its main advantage over metadata. The watermark is in the content itself and is designed to survive cropping, compression and re-encoding. Heavy editing or very low quality can still degrade detection, which is why the portal asks for the best copy you have.
Does ChatGPT or Midjourney use SynthID?
SynthID originated with Google and is used across its generative products, with adoption extending to several partners. The detector page lists which companies participate; check there rather than assuming, since the set changes.
Is a watermark proof that something is AI-generated?
A detected watermark is strong evidence that participating software produced the file. Its absence proves nothing, because most tools do not watermark at all.