Pumice · metadata scrubber
Every file remembers where it’s been.
Photos, videos, RAW files and PDFs quietly carry where they were made, the device and its serial number, the time, your name — and AI images carry the prompt and the generator’s signature. Pumice shows you all of it, then scrubs it out without changing a single pixel, frame or page.
Drop photos, videos, RAW files or PDFs
As many as you like. They are read here, in this tab, and never uploaded.
JPG · PNG · WebP · HEIC · AVIF · GIF · TIFF · MP4 · MOV · MKV · WebM · AVI · CR2 · CR3 · NEF · ARW · DNG · RAF · ORF · RW2 · PDF
What comes out
The details you never typed in, but every file carries.
Photos & AI images
- GPS position, altitude and the direction you faced
- Camera, lens and their serial numbers
- When it was taken, to the second, and the time zone
- Your name, copyright and the camera owner
- Prompts, seeds and models from Stable Diffusion, ComfyUI, Midjourney and others
- Content Credentials (C2PA) and IPTC “made with AI” labels
- Thumbnails, editing history and private app data
Kept: the colour profile, and the orientation — so phone photos don’t turn sideways.
Videos
- The location written into iPhone and Android videos
- Phone model, system version and recording time
- iPhone face-detection tracks, GoPro telemetry and drone GPS logs
- Titles, authors, comments and chapter names
- Content Credentials, XMP and editors’ leftovers
Kept: every video and audio sample, byte for byte. Nothing is re-encoded.
Camera RAW
- Body and lens serial numbers, including the ones hidden in maker notes
- Owner name, copyright and capture times
- GPS, and the metadata inside the embedded previews
- XMP, IPTC and the original file name inside DNGs
Kept: make, model and lens, which editors need to open the file — and the sensor data, checked identical.
PDFs
- Author, title, the app that made it and every date
- The hidden XMP copy of all that, and document IDs
- Earlier saved versions still inside the file
- Comment authors, digital signatures and their certificates
- GPS and camera details inside photos placed in the document
Kept: everything you can see on the pages, including comment text.
How it cleans
It doesn’t redraw your photo. It cuts the metadata out.
Most cleaners decode your photo and save a new one, which re-compresses it and quietly changes every pixel. Pumice works on the file’s structure instead: the blocks that hold metadata are removed or blanked, and every byte of the picture is copied across untouched.
Videos and RAW files are cleaned in place. A metadata block becomes padding of exactly the same size, so nothing moves, every offset stays valid, and a 4 GB video is cleaned without ever being loaded into memory.
Then the clean copy is read again from scratch, as if it were a new file. It has to show nothing identifying, and its picture, sound or pages have to match the original exactly — or you’re told.
What no metadata cleaner can do
- Anything you can see. A face, a street sign, a reflection, a visible watermark — it’s in the picture, not the metadata.
- Watermarks woven into the pixels. Some AI generators (Google’s SynthID, for one) hide a mark in the image itself. Removing metadata doesn’t touch it, and an AI detector looking at the picture can still recognise one.
- Copies already shared. Cleaning protects what you send from now on. A file you posted before still carries what it carried.
- Where you send it from. Services see your account, your IP address and when you upload, whatever the file says.
Your files are opened by this page, in your browser. There is no upload and no server copy: the page has no code that sends file contents anywhere. Cleaning runs in a background worker, so even a large video doesn’t freeze the tab.