About Remove AI Label
Remove AI Label is a free, browser-based image metadata inspector and stripper. It reads what is inside your photo, shows you the real values, and removes what you choose without re-encoding the picture.
It exists because the existing tools in this category were measurably failing at both halves of that job.
What the audit found
Before writing any code we audited the market leader. Its removal engine worked, but its detection layer only parsed C2PA manifests. Plain EXIF, XMP, IPTC and PNG text chunks were reported as "None" on all five of its reporting surfaces.
We tested it with a control file carrying GPS coordinates and an EXIF Software field reading "Adobe Photoshop 25.0 (Generative Fill)". It reported "Metadata percentage 0%" and told us the image was clean for upload. That file was neither clean nor safe to upload for anyone trying to avoid an AI label.
We also measured its removal. On a single pass it altered 52.3 percent of pixels, with a peak signal to noise ratio of 39.78 dB, and returned a PNG 15.8 percent larger than the input, while its homepage claimed the pixels were unchanged.
What we built instead
The engine parses EXIF including the GPS directory, XMP, IPTC, C2PA and PNG text chunks, and displays the actual field names and values it found. When it says nothing was found, that is because the parser ran and found nothing.
Removal is container surgery rather than decode and re-encode. Metadata lives in length-prefixed blocks beside the compressed image, so the file is copied forward with those blocks skipped. The compressed pixel data is never touched.
Every operation is verified afterwards. The output is parsed again, the pixel payload is compared byte for byte against the input, and the result is only reported as clean when that comparison, the size check and the zero-remaining-fields check all pass.
Where we say no
We do not inject fabricated camera EXIF to make generated images read as photographs. It is the least defensible feature in this category.
We do not claim to remove pixel watermarks, and we have a page explaining why nobody can.
We do not build pages for formats we cannot parse. A page ranking for "PDF metadata remover" that then tells the visitor to go elsewhere is a page that should not exist.
Each of those costs us traffic. They are the reason the rest of what we say can be trusted.
How it is built
The parsers are hand-written and dependency-free, which matters because the lossless guarantee depends on owning that code rather than trusting a library to behave.
Everything runs in a Web Worker in your browser. There is no server-side image processing and no upload endpoint. The site works with the network switched off, which is the most direct proof available that nothing is being sent anywhere.