What ChatGPT embeds when it generates an image
The typical trail is a signed C2PA credential, sometimes accompanied by XMP blocks and a reduced EXIF. It is OpenAI’s same provenance approach: the “generated by AI” label travels in the file’s metadata, not in the visible image.
How to clean it
- First the checker: it tells you whether your download carries C2PA and which other blocks it contains.
- Then the matching mode: /c2pa for the credential, /xmp and /exif for the rest.
- Download the clean file with no recompression; the pixels do not change.
Supported formats: JPEG, PNG, WebP, AVIF, TIFF and HEIC. If ChatGPT gave you a PNG or a WebP, that is fine.
What we cannot promise
We remove the label inside the file. We have no control over what OpenAI may log on its platform when the image is generated, and we do not remove any possible pixel-level mark (our work is on metadata and credentials, not on pixels). Be realistic: cleaning metadata reduces the visible trail in the file; it does not make it undetectable.
Privacy
The image is uploaded to our server, processed and deleted after 5 minutes. It is not used for training, we add no watermark, no account is required and the service is free.