Skip to content
Delete Image Metadata

Remove ChatGPT Image Metadata

Every ChatGPT image carries OpenAI's signed Content Credentials, and since May 2026 an invisible SynthID watermark too. Remove the first in your browser — and know that the second stays.

  • .jpg
  • .jpeg
  • .png
  • .webp
C2PA
OpenAI's signed credential, in the file's metadata
SynthID
Invisible watermark in the pixels since May 2026 — not removable
0
Pixels changed by removing the metadata
Processed in your browser
Ready

How it works

  1. 1

    Load the ChatGPT download

    PNG, JPEG or WebP, read in your browser; nothing is uploaded.

  2. 2

    See what is in the file

    OpenAI's C2PA manifest and any editor records are listed.

  3. 3

    Remove and verify

    The metadata is removed and the copy re-read. SynthID in the pixels is untouched.

Two markers, in two different places

Images made with ChatGPT carry C2PA Content Credentials: a manifest signed by OpenAI stating that the image was generated by its AI. It sits in the file's metadata, and it does not contain your prompt.

On 19 May 2026 OpenAI added a second marker. Images from ChatGPT, Codex and the OpenAI API now also carry SynthID, the invisible watermark Google developed, embedded in the pixels themselves. OpenAI published a research-preview tool, Verify, that checks an image for both.

What removing the metadata changes now

Removing the metadata takes away the C2PA manifest. Platforms that read credentials on upload — Instagram, TikTok, LinkedIn — find nothing in the file to read.

It does not touch SynthID, and no tool that leaves the pixels alone can. Before May 2026, stripping the credential left a ChatGPT image without a machine-readable mark; it no longer does. Anyone with a SynthID or OpenAI verifier can still identify where the image came from.

OpenAI's terms

OpenAI's Terms of Use do not let you represent output as human-generated when it is not. Removing the credential so that a generated image can be passed off as a photograph or as your own drawing does exactly that.

There are ordinary reasons to clean a file — keeping editing history and device details out of a published image, or reducing file size — and the remover handles those. What it cannot do is make a generated image anything other than generated.

Edits add more than OpenAI did

Saving a ChatGPT image, editing it and re-exporting usually adds more identifying metadata than OpenAI wrote: XMP edit history, document IDs, software names, and on a phone the device model. Those describe you and your tools rather than the generator.

Clean after the last edit, not before it, or the next export simply writes new records back in.

What this tool does

  • Shows OpenAI's credential and any editor records before you publish
  • Removes metadata losslessly, without re-compressing the image
  • States plainly what stays: the SynthID watermark

What it does not do

  • Cannot remove or detect SynthID, which is part of the pixels
  • Does not remove the copy of the image in your ChatGPT history
  • Does not change what OpenAI's terms say about presenting output as human-made

Remove ChatGPT Image Metadata

Every ChatGPT image carries OpenAI's signed Content Credentials, and since May 2026 an invisible SynthID watermark too. Remove the first in your browser — and know that the second stays.

Open the tool

Reviewed and updated . Behaviour described on this page is covered by the project’s automated tests.

Frequently asked questions

Does ChatGPT add metadata to images?

Yes. ChatGPT images carry C2PA Content Credentials signed by OpenAI, and since May 2026 they also carry an invisible SynthID watermark in the pixels.

Does a ChatGPT image contain my prompt?

Not in its metadata. The C2PA manifest states that OpenAI's AI generated the image; it does not embed the prompt text. Load a file above to see what yours holds.

Can this remove SynthID from a ChatGPT image?

No. SynthID is encoded into the pixel values, and this tool never changes pixels. Removing the metadata leaves the watermark exactly as it was.

Will Instagram still label a ChatGPT image after cleaning?

Instagram's documented trigger is metadata, which cleaning removes. But the image still carries SynthID, and any platform that checks for it can identify it, so no outcome is guaranteed.

Is it against OpenAI's terms to remove the metadata?

OpenAI's Terms of Use prohibit representing output as human-generated when it is not. Removing the credential in order to do that would breach them; cleaning editor records from a file you disclose as AI does not.

How is this different from the DALL·E page?

ChatGPT now generates images with OpenAI's newer GPT image models rather than DALL·E 3. Both carry OpenAI's credentials and SynthID, so the mechanics are the same; the DALL·E page covers older files.

Keep going

Related tools

Back to the main metadata remover