Where AI metadata hides in a JPEG, PNG and WebP, and how to confirm it is gone

How to read it, strip it, and confirm it is gone.

Same job, different file formats

AI generation records are small blocks of data stored beside the pixels. They are not visible in the picture, which is why people are surprised when an upload is labelled.

JPEG

A C2PA manifest sits in an APP11 JUMBF box. Content Credentials pointers are in APP1 XMP. The IPTC DigitalSourceType tag (for example trainedAlgorithmicMedia) is usually in that XMP.

PNG and WebP

A PNG uses a caBX chunk for the C2PA manifest and text chunks for a generator's recipe or prompt. A WebP uses RIFF chunks for XMP and C2PA.

Read, strip, read again

1. Read the exact file you plan to publish and note what is there.
2. Remove those blocks and write a new file.
3. Scan the new file's bytes again; a second scan, not a copy of the first list, is the proof.
4. Upload the cleaned file, not the original.

Do it in the browser

I built a free tool that does these three steps locally, so the image is not uploaded: an AI metadata remover that scans the cleaned file again. One image is free; a folder-sized batch is a paid option.

Limits

Metadata removal does not touch the pixels, so invisible watermarks such as SynthID stay, and a cleaned file does not force a platform to drop a label. Disclosure: I made this tool.# What your Etsy, Amazon and eBay listing photos can reveal before you upload them

A short checklist for online sellers.