Skip to content

ChatGPT · DALL·E · Sora

ChatGPT image metadata remover

Every image ChatGPT produces carries a signed record of how it was made. See exactly what is in yours, then remove it — without touching a pixel.

What ChatGPT embeds in an image

Since early 2024, images generated through ChatGPT and DALL·E 3 ship with aC2PA content credential. It is not a visible watermark and not a corner logo — it is a cryptographically signed manifest tucked inside the file, recording that the image was produced with OpenAI tooling and listing the operations applied to it.

The manifest lives in a JUMBF box. In a JPEG that is an APP11 marker segment; in a PNG it is a caBX chunk. Ordinary EXIF tools do not know those exist, which is why a generic EXIF cleaner will report a ChatGPT image as clean while leaving the credential exactly where it was.

Whatever handled the file afterwards may have added more. Download managers, converters, upscalers and editors each write their own Software and XMP history entries, so an image that has been through a workflow often carries a longer trail than the one ChatGPT originally produced.

What removal does, and what it does not

Removing the manifest deletes the signed history. A verifier then reports the file as carrying no credentials — the same state as any ordinary photograph that never had them. It does not forge anything: a signature can be present or absent, but it cannot be edited to say something false.

Three things it specifically does not do:

  • It does not make an image undetectable as AI. Pixel-level classifiers work on the picture, not the file wrapper. Invisible watermarking survives metadata removal completely.
  • It does not affect copies. Anything already published elsewhere keeps its credential.
  • It does not change the image. Only the container is rewritten; the pixels are copied through byte for byte.

Where we draw the line

This tool shows you what is in your file and removes it if you ask. It does not modify your pixels to defeat AI classifiers, and it will not, because that degrades your image in exchange for misleading a system built to label content accurately.

Worth knowing before you strip: most large platforms ask for AI-generated content to be disclosed, and the EU AI Act's Article 50 transparency obligations apply from2 August 2026. Cleaning metadata for privacy is a normal thing to do. Removing a disclosure so a client, a publication or a competition judge believes something false is not the same act, and it is judged differently.

Other AI generators

The same engine handles Midjourney, Adobe Firefly, Stable Diffusion, ComfyUI, Leonardo and the rest. Stable Diffusion interfaces are the most revealing of the group: they write your full prompt, negative prompt, seed, sampler, CFG scale and model hash into plain PNG text chunks that anyone can read in a text editor. The generalAI image metadata remover covers all of them, and if you have a whole folder of generations thebulk image metadata remover takes them in one go.

Questions

Frequently asked questions

What metadata do ChatGPT images contain?
Images generated by ChatGPT, DALL·E 3 and Sora carry a C2PA content credential — a cryptographically signed manifest recording that the file was produced with OpenAI tooling, along with the actions applied to it. Depending on how the image reached you it may also carry ordinary EXIF and XMP fields written by whatever downloaded, converted or edited it afterwards.
Does removing the metadata make a ChatGPT image undetectable as AI?
No, and you should not rely on that. Metadata is one signal among several. Platforms also run classifiers over the pixels themselves, and some providers embed invisible pixel-level watermarks that survive metadata removal completely. Removing metadata removes metadata.
Is it legal to remove C2PA credentials from an image I generated?
Removing metadata from a file you own is generally lawful. Whether you should depends on where the file is going. Most large platforms ask for AI-generated content to be disclosed, and the EU AI Act's Article 50 transparency obligations apply from 2 August 2026. Stripping provenance for privacy is ordinary. Stripping it so someone believes a generated image is a photograph is a different act.
Why would I remove it for a legitimate reason?
Provenance data is identifying data. A credential records the tool, the account context and a timeline. Photographers and designers delivering work to clients often prefer not to expose their internal workflow, and anyone publishing under a pseudonym has an obvious reason not to ship a signed record of how a file was made.
Does this work on images from other AI tools?
Yes. The same engine detects and removes Midjourney and Adobe Firefly tags, Stable Diffusion parameters stored in PNG text chunks — prompt, negative prompt, seed, sampler, CFG scale and model hash — and generic generator names left in the Software, UserComment or CreatorTool fields.