What AI-generated images hide
An AI image looks like any other picture, but the file often carries a layer of metadata that ordinary photos do not. Depending on which tool produced it and how you exported it, an AI image can contain:
- C2PA Content Credentials — a cryptographically signed provenance record stating the image was generated or edited by AI, which tool did it, and when. Major generators increasingly attach these by default.
- Generator tags in XMP or PNG text chunks — fields naming the model (for example, identifying a Stable Diffusion or DALL·E origin).
- The prompt and parameters — some Stable Diffusion workflows write the full text prompt, seed, sampler, and model name straight into the PNG's text chunks.
- Ordinary EXIF and XMP — added by any editor you passed the image through after generating it.
If you share an AI image without cleaning it, all of this travels with the file.
Why people remove it
There are several legitimate reasons, and we are not here to judge which applies to you:
- Privacy. You may simply not want to broadcast which tool you used, when, or with what prompt — the same way you would strip GPS from a photo.
- Prompt confidentiality. If your prompt is part of your craft or your business, the PNG text chunk that embeds it is an unintended leak.
- Clean re-use. When an image is one input into a larger design, the embedded provenance and parameters are just clutter you do not want carried downstream.
It is worth being straight about the trade-off: C2PA exists partly to support transparency about AI-generated content, and some platforms read it. Removing it is a deliberate choice. We default to removing it because most people sharing an everyday image are not trying to publish a verifiable provenance claim — but the decision is yours, and you should make it knowingly. Our C2PA guide covers the keep-versus-remove question in depth.
What this tool removes
- C2PA Content Credentials (the
caBXchunk in PNG; the equivalent in JPEG/WebP) - XMP packets, including generator and AI-origin tags
- PNG text chunks (
tEXt,iTXt,zTXt) that hold prompts and parameters - Standard EXIF, including any GPS or device data added by later editing
- Embedded thumbnails and timestamps
The image pixels are never re-compressed, so the picture you get back is identical to the one you put in — just without the hidden layer.
It runs entirely in your browser
This matters more for AI images than almost anything else, because the prompt and parameters can be genuinely sensitive. A tool that uploads your image to a server to clean it has seen your prompt. This tool does not: the entire process runs in JavaScript on your own device. You can verify it — open your browser's developer tools, switch to the Network tab, and watch as you clean a file: there is no upload. You can even disconnect from the internet after the page loads and it still works.
Frequently asked questions
Does this work on Midjourney, DALL·E, and Stable Diffusion images?
Yes. It works on the image files those tools produce (typically PNG, JPEG, or WebP), removing C2PA credentials, XMP and text-chunk tags, embedded prompts, and EXIF regardless of which generator made the file.
Will it remove the prompt embedded in a Stable Diffusion PNG?
Yes. Prompts and generation parameters are stored in PNG text chunks
(tEXt / iTXt), which this tool strips along with the other
metadata.
Does removing metadata change the image itself?
No. The pixels are not re-encoded. The visible image is identical; only the hidden metadata is removed.
Will this make my AI image undetectable as AI?
We make no such claim. Removing embedded C2PA and generator tags strips the declared provenance metadata, but visual AI-detection systems analyze pixels, not metadata, and are entirely separate. This tool removes hidden file metadata; it does not alter how an image looks or claim to defeat any detector.
Is anything uploaded?
No. Everything runs in your browser. Your image — and any prompt embedded in it — never leaves your device.
Can I clean a batch of AI images at once?
Yes. Drop multiple files and each is processed locally, then returned individually or as a single ZIP archive.
For the broader picture of image metadata, see our guide on what your photos tell strangers, or read about C2PA Content Credentials in detail.