How to Remove Watermarks from ChatGPT and DALL·E Images

Images you generate in ChatGPT (and via DALL·E) usually have no logo or visible mark — which is exactly why people assume they’re “clean.” They aren’t. OpenAI embeds C2PA metadata that identifies the image as AI-generated. Here’s how to check and remove it.

What’s hidden in a ChatGPT image

The marker is invisible to the eye but sits in the file’s metadata: C2PA Content Credentials recording that the image came from an OpenAI model. Upload such a file to a platform that reads Content Credentials and it can be flagged as AI-made automatically — no visible watermark required.

How to remove it

1. Drop your ChatGPT or DALL·E image into the free Unmarkr tool.
2. It detects the C2PA metadata and removes it, along with any EXIF/IPTC AI tags, without altering the pixels.
3. Download the clean file.

Want the detail on what those tags are? See how to remove C2PA and “Made with AI” metadata. And if you’re dealing with a tool that adds a visible badge instead, see removing the Gemini watermark.

Does ChatGPT use an invisible pixel watermark?

Today the main marker on OpenAI images is metadata, not a pixel-level watermark — which is why a metadata strip is enough. That’s different from Google, which can also embed an invisible SynthID signal in the pixels.