People talk about "the metadata" as if it were one block. In a file that has been through a camera, an editor and a publishing pipeline, it is usually three blocks built on different foundations, stored in different places, and read by different software.
That is the practical reason a stripped file can still leak. Delete the EXIF segment and the XMP packet is untouched. Delete that too and an IPTC caption may still name the photographer in APP13.
Three systems in one file
| System | Encoding | In a JPEG | Built for |
|---|---|---|---|
| EXIF | Binary TIFF-style IFDs | APP1, identifier Exif | Capture settings: exposure, lens, GPS, timestamps |
| XMP | RDF/XML text | APP1, Adobe namespace URI | Edit history, ratings, rights, cross-application properties |
| IPTC IIM | Binary datasets | APP13, inside a Photoshop resource block | Editorial captions, by-lines, credit and rights |
EXIF: binary, fixed and camera-first
An EXIF block is a TIFF file embedded in the image. It starts with a two-byte byte-order mark
— II for little-endian, MM for big-endian — then the number 42,
then an offset to the first image file directory.
An IFD is a count followed by twelve-byte entries. Each entry holds a tag number, a data type, a count and either the value itself or an offset to it. Values of four bytes or fewer are stored inline; anything larger lives elsewhere in the block and is reached by offset. Types include ASCII, SHORT, LONG, RATIONAL and their signed variants.
IFD0 holds the basic tags. Two entries point to sub-directories: 0x8769 for the
Exif IFD, which holds exposure and lens data, and 0x8825 for GPS. There is
often a third pointer to a thumbnail IFD, which is a second small JPEG with its own tags.
EXIF is compact, universally supported and terrible at evolution. The tag list is fixed by the standard, so vendors bolt on proprietary MakerNotes that only their own software reads. It has no concept of a namespace and no place to record an edit history.
XMP: XML, extensible and Adobe-shaped
XMP is a subset of RDF serialised as XML. Properties are grouped by namespace:
dc: for Dublin Core fields such as creator, rights and subject,
xmp: for dates and the creating tool, photoshop: for city, credit
and headline, crs: for Camera Raw develop settings. That extensibility is its
advantage: a new property needs no new binary specification.
The interesting namespace is xmpMM:, which holds provenance.
xmpMM:History is an ordered list of events, each with an action, a timestamp, a
software agent and sometimes the parameters used. xmpMM:DerivedFrom and the
document and instance identifiers let an application trace a file back through versions.
In a JPEG, the packet sits in an APP1 segment whose payload begins with the namespace URI
http://ns.adobe.com/xap/1.0/ followed by a null byte. A packet larger than the
space available is split: a small standard packet plus one or more extended-XMP APP1
segments carrying the remainder under a GUID.
IPTC IIM: the press record
IPTC IIM is a binary format built from datasets. Each dataset begins with the marker byte
0x1C, then a record number, then a dataset number, then a two-byte length.
Record 2 is the editorial record, and its dataset numbers are the ones that matter in
practice.
| Dataset | Field | Typical content |
|---|---|---|
| 2:05 | Object Name | Short title |
| 2:25 | Keywords | Repeated terms for search |
| 2:55 | Date Created | Editorial date, as eight digits |
| 2:80 | By-line | Photographer's name |
| 2:85 | By-line Title | Job title or role |
| 2:90 | City | Place of capture |
| 2:101 | Country | Country name |
| 2:105 | Headline | Publishing headline |
| 2:110 | Credit | Provider, not necessarily the author |
| 2:116 | Copyright Notice | Rights statement |
| 2:118 | Contact | Address of the rights holder |
| 2:120 | Caption-Abstract | The full caption |
In a JPEG, these datasets are wrapped in an APP13 segment starting with the identifier
Photoshop 3.0 and a null byte, then an 8BIM resource with ID 0x0404.
The nesting is why some tools report "no metadata" on a file that has a complete IPTC
caption: they never look inside APP13.
IIM is 8-bit and was designed before Unicode was universal, so non-Latin captions are often stored in a local encoding and display as mojibake. That limitation is a large part of why IPTC later defined IPTC Core and Extension as XMP schemas rather than extending IIM.
Why removing one leaves the others
The three systems are stored independently. Nothing in the format keeps them in sync, and no writer is obliged to update the others. A camera writes EXIF only. Lightroom writes XMP. Photoshop writes XMP and often IIM as well. A wire service may write all three.
Duplication is normal. The same caption can exist in dc:description in XMP and
in dataset 2:120 in IIM. The same coordinates can exist in the GPS IFD and as
exif:GPSLatitude in XMP. Deleting one copy changes nothing about the other.
There is also a container subtlety in JPEG. EXIF and XMP both live in APP1 segments, so a tool that removes "the APP1 segment" may remove one and leave the other depending on which it found first. Removing both requires finding both signatures, not just the first APP1.
PNG and WebP have the same problem in different clothing: a PNG can hold eXIf
and an XMP iTXt chunk, and a WebP can hold an EXIF chunk and an
XMP chunk. In every container, the count of metadata blocks is what matters,
not the count of metadata formats.
Removing all three
Work by signature, not by extension. In a JPEG, expect one or two APP1 segments, an APP13, possibly a COM, and an APP2 if there is a colour profile. In a PNG, walk the chunk list and read every text chunk's keyword. In a WebP, check the VP8X flags and then the chunk list.
If you are weighing which system to keep for legitimate reasons: EXIF for capture settings, XMP for workflow and rights in Adobe tools, IIM when a downstream agency requires the old record.
For the tag-level view, see what EXIF contains, and for the containers see JPEG segments and PNG chunks. The next guide covers the two blocks that should stay.
See what your photos are carrying
Drop an image in. MetaPurge lists every tag, then rewrites the file without touching a pixel.
Open the metadata removerWhere to go next
- ICC profiles and orientation — the two blocks a strip should preserve.
- Removing metadata from JPEG — where each of these blocks sits in the segment list.
- Removing metadata from WebP — the RIFF equivalent.