AIWatermarkCheckerOpen tool

OpenAI-focused

GPT image detector

Drop an image and read its provenance: OpenAI's image generations carry embedded C2PA metadata — a signed “this was made by this model” record. This page decodes it in your browser, with zero uploads.

Runs entirely in your browser — your file is never uploaded.

Drop an image here, or browse files

JPEG · PNG · WebP — up to 25 MB. Files are read locally and never uploaded.

Try a sample:

Quick answer

GPT Image / DALL·E outputs carry signed C2PA metadata naming their generator. If an image historically came from ChatGPT and never got re-saved or compressed, that record usually survives — but screenshots, messenger compression, and social-platform processing commonly remove it. A “no provenance found” verdict therefore does not mean the image is human-made.

What the metadata looks like

A healthy OpenAI-style render shows:

Don't rely on vendor names alone: we treat generator strings as a hint family, and results name exact strings so you can judge (see the evidence table under the tool).

Where provenance dies (a survival table)

EventC2PA metadata
Saved / copied as-isusually intact ✔
Re-saved by an editor without credentials supportusually lost
Screenshotgone
Platform upload (most social/messaging apps)frequently stripped or re-encoded

FAQ

How can I tell if an image was made by ChatGPT / GPT Image?＋

OpenAI embeds C2PA provenance metadata in image generations (first announced for DALL·E 3 in February 2024 and carried to GPT-4o / gpt-image outputs). This tool reads that metadata in your browser: if it's intact you'll see the generator claim and CreationRecord-style actions.

Does the C2PA metadata survive screenshots or re-uploads?＋

Screenshots, image compression pipelines, and many social-media processors strip or break the metadata. A clean result here is common for images that were once labeled. Absence of provenance is not proof of human authorship.

Can the metadata be faked?＋

C2PA claims are cryptographically signed; someone without the vendor's signing infrastructure cannot forge a credible generator claim. (This reader displays the signature box; full certificate-chain validation is on our roadmap — see methodology.)

What about OpenAI's SynthID-style invisible watermark in images?＋

OpenAI's Verify page lists SynthID among the evidence its verification uses. SynthID pixel detection runs on Google-provided verification gates, not independent third-party readers, so this page reads the open C2PA layer.

Is this tool free and private?＋

Yes on both. There's no account, no upload — all parsing happens in JavaScript on your device.

Sources