DALL-E image watermark detector

The standalone DALL-E tool, ChatGPT's image generation, and today's C2PA metadata each leave different evidence. Check which one your file has right here.

Use the toolSee what it catches →
Image watermarks · dedicated media workflow
JPEG · PNG · WebP
Drop your images hereCheck your images for AI watermarks and content credentials: free, no sign-up.Open the ChatGPT image watermark detector

This opens the dedicated ChatGPT image watermark detector. You upload there, and detection, cleaning and download all happen in one place. Files are never published or shared.

of watermark history
3 eras
to a finished check
3 steps

DALL-E watermark detector

What this DALL-E image watermark detector does

DALL-E has left different evidence at different points: a visible signature on the original standalone tool, a corner mark on some ChatGPT-era images, and a C2PA metadata manifest more recently. This app checks a file against all three before telling you which one applies.

Drop your image into the upload control above and it opens our media workflow, which loads the file and checks it against each of these three signals in the same session.

The problem

Checking a DALL-E image for a watermark, in plain terms

The original dalle.com tool added a small multicolor signature to a corner as a standard part of the output, nothing hidden about it. Once image generation moved inside ChatGPT, that specific signature stopped being the default, and the evidence shifted toward whatever OpenAI's broader image-marking approach was doing at the time.

So the honest answer changes depending on which DALL-E you mean. A file from the original tool and a file generated in ChatGPT today can carry completely different kinds of evidence.

The workflow

How to check a DALL-E image for its watermark in three steps

What matters is the actual file in front of you, not a general rule about DALL-E.

1. Get the original download
Not a screenshot. A download can carry a C2PA manifest and keeps a visible signature or mark at full quality if one is present.
2. Drop it into the checker above
The upload control opens our media workflow, where the manifest is actually read and any visible mark is confirmed.
3. Match the result to the era
A pre-ChatGPT DALL-E 2 file, a ChatGPT-era file, and a current file can each carry a different mix of evidence.

Why it matters

What a missing watermark actually tells you

A visible signature or a verified manifest is solid evidence, hard to fake convincingly after the fact. Finding either one tells you a fair amount about which tool, and often which era, made the file.

Finding nothing tells you far less. Cropping removes a visible signature outright, and metadata gets stripped by screenshots, re-saves and most social platforms as a matter of routine. Read an empty result as unresolved, not as proof.

The mechanism

Why DALL-E's evidence looks so different across eras

DALL-E began as a standalone product with its own distinctive visible signature, then became a generation path inside ChatGPT, folded into OpenAI's broader and changing approach to image marking. Each phase reflects a different product decision, not a single consistent rule.

It's also a different category of signal from Google's SynthID, an in-pixel statistical mark rather than a visible logo or a metadata record. Across every era, DALL-E's evidence has lived somewhere you can see or somewhere a reader can check, never somewhere invisible to both.

Straight answer

Does this prove an image came from DALL-E?

A verified signature or manifest is strong evidence for that specific file. It doesn't extend to every DALL-E image lacking one, since cropping and ordinary sharing remove both kinds of evidence all the time without anyone hiding anything.

Treat a result as one input alongside what you already know about where the file came from, not a single answer that closes the case. Weigh a positive finding more heavily than an empty one, since so much ordinary sharing produces an empty result on its own.

Who uses it

Who checks a DALL-E image for a watermark, and why

A few recurring, practical situations.

Newsroom verification
Checking a circulating image for signs of DALL-E and roughly which era it's from before it runs in a story.
Platform trust and safety
Reading a visible signature or C2PA metadata as one signal within a broader policy.
Stock and licensing review
Confirming whether a submitted file declares a generative tool chain.
Anyone checking one file
Running a single image through the check out of plain curiosity about which era made it.
FAQ

Everything answered. Everything people ask about the DALL-E image watermark detector

Does DALL-E add a watermark to its images?

It depends on the era. The original standalone tool added a visible signature by default; images from ChatGPT have carried a corner mark on some releases and C2PA metadata more consistently.

How do I check whether an image came from DALL-E?

Get the original download and drop it into the upload control on this page, which runs the check.

Is my file uploaded when I use this?

Yes, to our own media workflow, so it can actually be read. It isn't published or shared.

What did the classic DALL-E watermark look like?

A small multicolor square signature placed in a corner, a visible feature of the original standalone dalle.com tool.

Is that the same mark as ChatGPT uses?

Not necessarily. Once DALL-E moved inside ChatGPT, the evidence shifted toward OpenAI's broader and evolving image-marking approach.

Is a screenshot as good as the original download?

No. A download can carry a C2PA manifest and keeps a visible mark at full quality. A screenshot reliably carries neither.

Does cropping remove the evidence?

Cropping removes a visible signature or mark specifically. It has no effect on metadata, which sits in the file's data, not the visible picture.

Can the C2PA metadata be forged?

The signature behind it is cryptographic, so producing a convincing fake is difficult.

Is this app made by OpenAI?

No. This is an independent detector with no affiliation to OpenAI.

How is this different from the ChatGPT image detector?

They overlap closely. This page leans on DALL-E's own history, including the standalone tool's signature, which predates ChatGPT's image generation entirely.

Is there a check limit?

Ordinary photo files check without trouble. A very large file just takes a little longer.

Does this cover video too?

No, this route is for images. Video watermark checks have their own route on this network.

If neither signal shows up, is the image confirmed real photography?

Not reliably. Both can go missing through ordinary handling rather than because the file never touched DALL-E.

Can I check a file I don't have the rights to reuse?

Checking provenance is separate from reuse rights. Confirm you have the rights to whatever you actually do with the file.

Is the checker free?

Yes. There's no cost to run a check through this app.

Does it work on mobile?

Yes, both the upload control and the check work in a phone browser.

Does this predict a third-party AI detector's result?

No. It reports what's actually in the file, not what a separate detection service might say.

How do I tell which era my file is from?

Check how it was generated. Anything from the standalone dalle.com site predates ChatGPT's image generation; anything made inside ChatGPT is the later era.

Does this cover Bing Image Creator, since it's also DALL-E-based?

No. Bing runs its own separate deployment with its own undocumented behavior, covered on its own dedicated route.

Can a single image carry more than one of the three signals?

Yes. A ChatGPT-era image can carry both a corner mark and C2PA metadata at once, depending on the release, so the checker looks for both rather than stopping at the first one found.