When you save an image from ChatGPT, Gemini, Midjourney or a Stable Diffusion setup, the picture is not the only thing in the file. Tucked into the bytes around it is a small record of how the image was made: a C2PA tag marking it as AI-generated, often the exact prompt and seed you used, and sometimes ordinary EXIF fields. This guide shows what is in there and how to strip it before you share the image.
What AI tools write into an image
The hidden data falls into three buckets.
Content Credentials (C2PA). C2PA is a provenance standard that several large companies have adopted. It writes a signed record into the file stating that the image was produced or edited by AI and naming the tool. Adobe shows it as Content Credentials; ChatGPT images carry it too. It does not appear on the picture, but any app that reads C2PA can display it.
Generation details. Many tools store the prompt, the negative prompt, the seed, the model, and the sampler right in the file. Stable Diffusion and ComfyUI write this into PNG text fields, where anyone who opens the file in a text viewer can read your whole prompt.
Plain EXIF. On top of the AI-specific tags, an image can still carry the everyday metadata any photo holds: software name, timestamps, and occasionally a description.
Why you might want it gone
There are a few ordinary reasons to clear this data, none of them dramatic.
You may not want to hand over the prompt that produced a result you spent time refining. You may be submitting work to a platform that treats a C2PA tag as a reason to label or downrank a post, and you want the image judged on its own. Or you simply prefer to share a clean file, the same way you would strip GPS from a holiday photo before posting it.
How to see what is buried
Before stripping anything, it helps to look. Open the AI metadata remover and drop an image in. It reads the file on your device and lists what it finds, so you can see whether the picture carries Content Credentials, a stored prompt, or EXIF fields. Checking first tells you whether a given image even needs cleaning, since not every export carries the same tags.
How to remove it
To clean an image with the AI metadata remover:
- Add your JPGs or PNGs. You can drop a whole batch at once.
- Read the list of what was found in each image.
- Click to remove the metadata and download the clean copies, singly or as one zip.
The clean files keep the picture exactly as it was. The C2PA chunk is dropped, the prompt and seed go with it, and the EXIF fields are cleared. You can re-open a cleaned file in the metadata remover or the EXIF viewer to confirm the tags are genuinely gone rather than just hidden from view.
What stripping metadata does not do
It is worth being clear about the limit, because the word “watermark” gets used loosely. Removing metadata clears the tags written into the file. It does not change the pixels, so two things survive:
- An invisible pixel watermark such as Google SynthID, which is woven into the image data itself. A detector that reads pixels can still flag the image after the metadata is gone.
- A visible logo drawn into the corner of the picture, like the Gemini sparkle. That is part of the image and has to be cropped or edited out separately.
If your goal is a clean file without the provenance tags and the stored prompt, metadata removal handles it completely. If you were expecting an invisible AI fingerprint to disappear, no metadata tool can do that, and any that claims to is removing the visible parts only. The honest version of “remove the AI watermark” is covered in the companion guide on what actually works.
