AI Watermarks and Content Credentials

Three different systems claim to mark AI images: SynthID, C2PA Content Credentials, and invisible pixel watermarks. They work differently, they fail differently, and only some of them can be checked by anyone other than the company that made them.

✓ Free✓ No sign-up🔒 Deleted after analysis

The short version. Content Credentials are readable by anyone but easy to strip. SynthID survives almost anything but only Google can read it. The Stable Diffusion watermark sits in between. None of them is present on most AI images, which is why a pixel-level detector is still the check that always works.

⬆️

Drag & drop, paste, or click to upload

PNG · JPG · WEBP · HEIC — your image is analyzed privately and deleted

No image? Try one:

Analyzing your image…

Likely realLikely AI-generated

This is an estimate, not proof — the detector can be wrong, producing both false positives and false negatives. Don't rely on it as your only source; verify anything important. wedetect.ai is not liable for decisions made from these results. See our Terms.

Inspect the metadata →

What we can and cannot check

Mark	Where it lives	Survives re-saving?	Can we check it?
C2PA Content Credentials	Signed metadata	No — platforms strip it	Yes, and we verify the signature
Stable Diffusion invisible watermark	Pixels	Yes, including screenshots	Yes
IPTC / XMP AI tags	Metadata	No	Yes
Google SynthID	Pixels	Yes	No — only Google can

We say this plainly because several tools imply they detect SynthID. They cannot. Google has not published a public verifier, so any third-party claim to read it is unfounded.

SynthID — Google's invisible watermark

SynthID embeds a pattern directly into the pixels of images, audio and video produced by Google's AI tools. Because it is in the pixels rather than the metadata, it survives cropping, resizing, compression and screenshots, which is exactly where Content Credentials fail.

The trade-off is that verification is closed. Google detects SynthID with its own model; there is no public verifier, so neither we nor any other independent checker can confirm a mark is there or rule one out. If you need a SynthID answer specifically, Google's own tooling is the only route.

C2PA Content Credentials — the signed record

Content Credentials are a manifest attached to the file and signed with a certificate. They record which tool or camera created it, what actions were applied, which source images were used, and who signed it. Because the manifest is signed, altering the image or the record breaks the signature, and a checker can tell you that it broke.

This is the richest provenance available, and the most fragile. It lives in metadata, so any platform that strips metadata removes it, which most do. A missing manifest is the normal case, not a red flag.

Check any file with the Content Credentials checker, which shows the signer, whether the signature validates, the recorded actions and whether the file declares AI generation or a camera capture.

Which tools attach what

Tool	Content Credentials	Invisible watermark
OpenAI image models	Yes	No
Adobe Firefly / Photoshop	Yes	Yes
Google (Imagen, Gemini, Nano Banana)	Yes	SynthID
Microsoft Designer	Yes	No
Stable Diffusion (official library)	No	Sometimes
Midjourney	No	No
Flux	No	No
Grok (Aurora)	No	No

Per-generator detail, including how each one behaves once a file has been through a platform: which AI generator made this?

Why none of this replaces a detector

Every marking scheme shares one weakness: it only works if the tool that made the image chose to participate. Midjourney, Flux, Grok and most community Stable Diffusion setups mark nothing at all, and a screenshot of any AI image discards every metadata-based signal in one step.

That is why provenance and pixel analysis are complementary rather than competing. A credential, when present, is the strongest evidence there is. When it is absent, which is most of the time, the AI image detector measures the pixels and still returns an answer.

Frequently asked questions

What is SynthID?

SynthID is Google's invisible watermark. It is embedded into the pixels of images, audio and video made by Google's AI tools, and it survives cropping, resizing and re-saving. It can only be verified by Google, through its own tools, so no third-party checker including ours can confirm or rule out a SynthID mark.

What are Content Credentials?

Content Credentials are a cryptographically signed record of how a file was made, defined by the C2PA standard. They name the tool or camera, list the actions applied, and break if the file or the record is altered. OpenAI, Adobe, Google, Microsoft and newer Leica, Nikon and Sony cameras attach them. Unlike a watermark they live in the file's metadata, which means they can be stripped.

What is the difference between a watermark and Content Credentials?

A watermark is hidden in the pixels and survives re-saving but usually cannot be read by anyone except the tool that made it. Content Credentials sit in the metadata, can be read and verified by anyone, and prove far more detail, but any platform that strips metadata removes them. They solve the same problem from opposite directions.

Can you check an image for an AI watermark?

Partly, and it depends which one. We detect the Stable Diffusion invisible watermark and we read and verify C2PA Content Credentials. We cannot verify Google's SynthID, because only Google can. No honest third-party tool claims otherwise.

Does a missing watermark mean an image is real?

No. Most AI images carry no watermark at all, most tools do not attach Content Credentials, and every major social platform strips metadata on upload. Absence proves nothing. That is exactly why a pixel-level detector still matters.

Which AI tools add Content Credentials?

OpenAI's image models, Adobe Firefly and Photoshop, Google's tools and Microsoft Designer attach them. Midjourney, Stable Diffusion in most community setups, Flux and Grok generally do not. The table above is kept current, and you can check any file yourself with our Content Credentials checker.