Three kinds of hidden labels in one file

Photo metadata is text and numbers stored inside the image file alongside the pixels, and it comes in at least three formats: EXIF, XMP and IPTC. They overlap in purpose, differ in structure, and sit in different parts of the file, which is why a tool that clears one can leave another untouched.

Why "metadata" is not one thing

People say "the metadata" as though the file had a single label on it. It has several, written by different programs at different times for different audiences. The camera writes one set when you press the shutter. An editing app may add a second when you adjust the picture. A photo agency, a newsroom or a cataloguing tool may add a third to say who owns it and what it shows.

Each set follows its own specification and lives in its own container inside the file. That matters for a reason that has nothing to do with taste. When you want to know what a file reveals, or what a service could learn from it, you need to know which containers exist, because a reader that parses only one will report a clean file while the others sit there. This is the starting point for everything on this site about what a photo carries and how to remove it before uploading. Neither of those pages is repeated here. This one is the map.

The file as a container

A JPEG is not a single block of pixel data. It is a sequence of marked segments. Some segments hold the compressed image. Others, called application segments and numbered APP0, APP1, APP2 and so on, are reserved for extra information that is not pixels. Different metadata formats claim different application segments, or different slots inside them.

Here is a simplified picture of a typical JPEG from a phone.

[ start of file ]
[ APP1  : EXIF block      - camera, time, GPS, orientation, thumbnail ]
[ APP1  : XMP packet      - XML text, edits, ratings, keywords        ]
[ APP13 : IPTC block      - caption, creator, rights, keywords        ]
[ ...tables describing how the pixels are compressed...               ]
[ compressed pixel data                                               ]
[ end of file ]

The exact order varies and a given file may have only some of these. The point is that they are separate blocks. Deleting one does not delete the others, and the pixel data is a different thing from all of them. Other formats rearrange the layout. PNG uses named chunks, HEIC uses a box structure derived from a different container design, and TIFF-based files use directories of tags. The three metadata formats still appear, in different wrappers, which is why HEIC and JPEG handle the same fields differently and PNG can carry text chunks that JPEG-oriented advice never mentions.

EXIF: what the camera knew

EXIF stands for Exchangeable Image File Format. It was designed for cameras, and it records what the device knew at the moment of capture. That is the key to remembering it: EXIF is machine-written and mostly automatic.

Typical contents:

• Camera make and model, and often the lens.
• Exposure settings such as shutter speed, aperture, ISO and focal length.
• Date and time of capture, often with a time zone offset.
• Orientation, a small flag telling software which way up the picture goes, which affects how a model sees the image.
• GPS coordinates, if location was enabled, in a block of their own.
• Sometimes a device or sensor serial number, as covered separately.
• A small preview image, the embedded thumbnail, which can outlive an edit.
• A "maker note", a vendor-specific region whose contents the standard does not define.

Structurally, EXIF is a set of numbered tags organised into directories, a design inherited from the TIFF format. It is compact and binary, designed to be quick to write inside a camera with little memory. Readers need to know the tag numbers to make sense of it, which is why a plain text editor shows gibberish and a metadata viewer shows names. EXIF is also the part most people mean when they say "my photo has my location in it", and it is the most common thing a privacy tool removes.

XMP: the extensible text layer

XMP, the Extensible Metadata Platform, was introduced by Adobe and is built on XML. Where EXIF is a fixed binary table, XMP is a text document embedded in the file, structured as properties grouped into namespaces. A namespace is a named vocabulary, so different communities can define fields without colliding with each other.

That flexibility is the main difference. EXIF has a defined list of tags. XMP can hold any property someone writes a schema for. In practice an XMP packet often records:

• Edit history and the name and version of the software used.
• Star ratings, colour labels and keywords added in cataloguing tools.
• Face or region tags, if a photo manager wrote them.
• Copies or restatements of EXIF values, since many programs mirror fields into both places.
• Rights and creator fields, using the IPTC vocabulary described below.

The mirroring is where people get caught. Software that rewrites EXIF but ignores XMP can leave the same date or camera name sitting in the text packet. A tool that reads XMP first can report details the EXIF block no longer contains. This is one reason editing a photo does not reliably clean it: the editor may update one layer and carry the other forward, because photographers want that information kept.

An XMP packet is also readable by eye, because it is XML text, which makes it the layer most worth looking at if you are inspecting a file by hand.

IPTC: the editorial layer

IPTC refers to the International Press Telecommunications Council, a standards body from the news industry. Its Photo Metadata Standard is described by the organisation as a schema for administrative, descriptive and rights information about pictures. It exists because news and stock agencies needed a shared way to say who made an image, who owns it, what it shows and under what terms it can be used.

So IPTC fields are about people and publication rather than devices:

• Caption or description.
• Creator or photographer name, and credit line.
• Copyright notice and usage terms.
• Keywords and subject codes.
• Location as words, such as a city or place name the photographer typed, as opposed to GPS coordinates.
• Contact details of the rights holder.

There are two things to untangle. Historically IPTC data was stored in its own older block, the Information Interchange Model, inside a Photoshop-style application segment of the file. The modern IPTC schemas are expressed as XMP properties. So a current file might carry IPTC information in the XMP layer, an older file in the legacy block, and a file processed by several generations of software might carry both.

For most casual phone photos, IPTC fields are empty. They fill up when a photo passes through professional or semi-professional tools, or when a photo manager lets you add captions and keywords. Empty is not the same as absent, and the fields are worth checking for exactly that reason.

The three side by side

The columns are tendencies rather than rules. Any program can write almost anything into XMP, and some cameras write XMP directly. The table is a way to guess where to look, not a guarantee of where something is.

Why the differences matter for removal

If these were one thing, one deletion would clear them. They are not, and three practical consequences follow.

First, removal is per container. A tool that clears EXIF has not necessarily cleared XMP or IPTC. A thorough tool enumerates the segments and chunks and removes the ones that carry descriptive data, which is the approach behind stripping metadata fields, as opposed to re-drawing the picture into a fresh file. ExifTool, the widely used command-line reader and writer, documents that it does not modify the image data itself when editing tags, and warns that removing all metadata is not appropriate for every file type, since some formats such as RAW images depend on it. Both points reinforce that clean removal is a deliberate operation, not a side effect.

Second, readers disagree. A viewer that shows only EXIF will call a file "clean" when the XMP packet holds a name. Before deciding what a file reveals, use a reader that lists every block, as the walkthrough on reading metadata yourself describes.

Third, metadata is not the only thing a file carries about its origin. The pixels themselves have properties that survive every deletion above, a subject taken up in the post on sensor noise. Metadata removal addresses labels. It does not address the picture.

What a service can and cannot do with it

When you upload a file, the server receives all the containers unless something upstream removed them. Whether the service reads them is a design choice. A tool that only needs the pixels can discard the metadata immediately on arrival, which is data minimisation in practice. A tool that keeps the original file intact keeps every label, and any later reader of that stored file, including staff, a breach or a processor, gets them too. Which of those a given service does is a question for its documentation, not something you can read from the interface.

Some layers are also read on purpose by a scoring pipeline. Orientation, as noted, is applied to rotate the pixels. Colour profile information, stored alongside the metadata, controls how colours are interpreted before inference, a topic in colour handling before the model. So not all metadata is incidental: part of it is instructions to the software that decodes the image, and removing it carelessly can change what the model sees. That is the accuracy-side reason to understand the formats rather than delete blindly, and the privacy-side reason to delete the right things.

Metadata that did not exist a few years ago

The list above is not closed. New provenance schemes are being embedded in photos, covered in content credentials and provenance tags. Some cameras now write richer device information. AI generation tools have written their prompts into PNG text fields. Each of these arrives as a new block or a new namespace, which is exactly the situation XMP's extensibility was designed for and exactly why a fixed list of fields to strip ages badly. The durable habit is to inspect the actual file rather than trust a checklist, and to prefer approaches that drop everything non-essential rather than naming each thing to remove.

Where this leaves you

You can hold the whole subject in one sentence: a photo file is pixels plus several separately stored label sets, and each label set has its own author, format and fate. EXIF is what the device knew. XMP is what software added and what travels in text. IPTC is what a person or agency declared about ownership and content.

If you want a reference for when someone sends you a photo and you wonder what it says about them, the reading step comes first and the removal step second. Practical etiquette for what to do with someone else's file is covered at ratepenis.com, and the photo-taking advice that makes cleaner originals in the first place is at penisrater.com. If your interest is in what data gets collected, kept and reasoned about rather than in the file format, measuremycock.com treats the data side of measurement. What one particular product does with the file on arrival is product documentation rather than general knowledge, and for Rate Cock it starts with its own scan explainer. Whatever any service does, the file you upload arrives with whatever your phone wrote, so the habit of checking is yours to keep.

One detail is worth ending on, because it surprises people. The most identifying thing in a typical photo file is often not the location. It is the combination of ordinary-looking fields, a device model, a timestamp, a software version and a serial, none of which looks personal on its own and all of which together can single out one phone.