About
A file can hold two stories
Last updated 22 November 2026
One is the picture, which is what you meant to send. The other is a set of notes about how it was made, which the camera added and most people never see. This page reads the second one out loud, then hands you a copy with it removed.
What a metadata block is
A JPEG is not one object. It is a sequence of segments with a picture at the end, and the format reserves the early segments for information the decoder does not need to draw anything. EXIF lives in APP1: a two byte byte order mark, a magic number, then a tree of directories, each holding entries of an id, a type, a count and a value. The value may be a number, a string, or a pointer to another directory. That tree is why a camera can write a shutter speed, a lens model and a coordinate without touching a single pixel.
XMP is the newer one, plain XML in its own APP1 segment, and it often repeats the EXIF fields in text plus a few of its own: a title, a description, a city. IPTC sits in APP13 and comes from the picture desk side of the trade, with a headline, a byline and a copyright line. A comment segment is free text that anyone could have written.
How the reader walks it
For a JPEG the reader steps from marker to marker, reading each segment's length, and stops when it reaches the scan marker, which is where the compressed image begins. For a PNG it reads a signature and then a chain of chunks, each with a length, a four letter type and a checksum, and inflates the compressed text chunks so a zTXt is not left as noise. For a WebP it reads the RIFF header and the chunks that follow, where EXIF and XMP sit beside the image as equals. All three hand the same TIFF reader the same kind of block, which is why a camera's tag names read the same whatever container they arrived in.
Values are printed rather than dumped. A rational of 1 over 250 becomes 1/250 s, a flash bitmask becomes words, an orientation of 6 becomes rotated 90 clockwise, and a timestamp in the camera's colon notation becomes something a person can read. Tags the reader does not know are still shown, with their hex id and their type, because a value you cannot name is still a value the file carries.
How the remover works
The first mode is subtraction. The file is walked again, and the segments or chunks named in the options are left out of the output while everything else is copied byte for byte. A JPEG keeps its own scan data untouched, a PNG keeps each chunk's own length and checksum so the rest stays valid, and a WebP gets its RIFF size corrected and its feature flags cleared so a reader does not go looking for a block that is no longer there. The picture cannot change, because the picture was never decoded.
The second mode is the one every image editor uses: decode, draw to a canvas, encode again. That removes everything, including blocks this page does not know about, at the cost of a small generation loss and any colour profile the original carried. Both results are read back with the same parser before the download link appears, so the report under it is a measurement rather than a promise.
What the samples are
The two sample buttons draw a landscape with a canvas and then build the metadata by hand: a TIFF tree with a camera, a lens, an exposure and a GPS directory, an XMP packet, a comment, and for the PNG a set of text chunks and an eXIf chunk. The camera is a fixture, the city is invented, and the coordinates land in an empty stretch of desert. Nothing in either sample belongs to a person, a device or a place, which is the point: you can inspect the reader without putting your own photographs into it.
What it leaves to you
- Whether the notes need to go. A portfolio shot may reasonably keep its camera and lens; a photo of your flat sent to a stranger should not keep its coordinate.
- Which mode to use. Subtraction is exact and keeps the picture identical. Re-encoding is the only way to be sure about blocks this page does not recognise, and it costs a little quality.
- What the picture itself shows. A street sign, a reflection, a window with a light on. Cleaning removes what the file says, not what the frame says.
- What happens after the download. The clean copy is a normal file; the original, if you kept it, still has everything in it.
- Whether the platform strips it anyway. Many large sites do, some do not, and most keep the original server side.
Who made it
Pikkola Lab builds small tools that keep the work in the browser. This is tool No. 58. Write to hello@pikkola.com with a file whose metadata this page read wrong, and say which camera wrote it.