Photo metadata after upload: a six-step round trip test

For most categories on this site, the data worth keeping is a record about the service. For photo services it is the other way round: the data is the pictures, and the service is a place they are temporarily displayed. Getting that the wrong way round is how people end up with an archive that is a folder of thumbnails.

This page covers photo platforms data in that sense: the fields inside a photo beyond the image itself, named by their standards (Exif, IPTC, XMP and the ICC color profile).

It explains what a service does on the way in and out, plus a six-step round trip test that shows whether what you get back matches what you put in, for whichever service you use.

What to take away

What is inside a photograph

An image file is pixels plus a set of embedded fields, and the fields carry more than people expect.

Capture data. Camera, lens, exposure, and often the date, time, and coordinates. This is the Exif block, written by the camera or the phone without being asked.

Rights and description. Creator, credit line, copyright notice, caption, keywords. These are written by you or by your editing software, and they are the fields that carry attribution when an image travels. They are the IPTC fields, named after the International Press Telecommunications Council, which developed its Information Interchange Model in the early 1990s for exchanging news.

Color information. A profile describing how the numbers in the file map to colors, usually an embedded ICC profile. Strip it and the image looks different, usually flatter, on some screens and not others.

Everything a service adds. Some platforms write their own fields on export, typically in the XMP block. The Extensible Metadata Platform has largely superseded the old IPTC file structure, but the IPTC attributes are defined in the IPTC Core schema for XMP, so a file can carry the same credit line twice, and a service may update one copy and not the other. It is worth knowing what is being added to your file in your name.

Two of those four matter for very different reasons.

The rights fields are the only attribution that travels with a file, so losing them costs you credit. The capture fields can include a location, so keeping them can publish where you live. Both failure modes are common and pull in opposite directions, so control the fields deliberately rather than prefer stripping or keeping in general.

What a service does on the way in and out

Every service re-encodes. That is not a criticism: an uploaded file has to be turned into several display sizes, and doing so requires decoding and re-encoding, which for the ordinary photographic formats is lossy and cannot be undone. The questions worth answering are what is kept alongside that.

What that costs is easiest to see when it is exaggerated: the same picture, encoded harder and harder from one side to the other, until detail turns into flat blocks. A single upload is nothing like that severe. An upload, a save from the page, and a re-upload three years later are cumulative, and nothing puts the detail back.

The filename row looks trivial and is not. An export of several thousand images with machine-generated names and no folder structure is technically complete and practically unusable, and the fix afterwards is days of work. Checking the naming and structure of what comes back before you commit is a ten-minute job.

The round trip test

Three images, chosen to stress different things. One with rich embedded fields and a known caption. One at the largest size you actually shoot. One with fine detail and gradual color, which is where re-encoding shows. Results differ by service and change without notice, so this page gives the test rather than a table of named services: run it on the one you use and the answer is about your files.

1. Read the embedded fields of each original with a metadata reader and save the output. ExifTool is a common choice: a command-line application for reading, writing and editing metadata that handles Exif, GPS, IPTC, XMP and ICC profile data.

The same six steps work for any service, whether it is a photo host, a cloud photo library or a social network's own upload, because they read the files rather than the service's description of itself.

Write the result down with the date. Services change this behavior without announcement, so the test is a dated observation rather than a permanent finding, and it belongs in the record the same way any other observation does. The general habit of storing the observation rather than a bare value applies here without modification.

What to keep, and where

The rule that survives every change in this category: the service holds a display copy, and you hold the archive. Whether you use a photo service for display, sharing or storage decides how much else you need, and none of them changes this rule.

This is the same argument as the pipeline for holding your own copy of a service's numbers, applied to files rather than figures, and it is easier here because the files are already yours.

The claims to be careful with

"Original quality" means whatever the service defines, often on a pricing page rather than in the help center. "Unlimited" storage generally applies to a specific processed format, not to whatever you upload.

"Backup" in a product name is a marketing word, not a commitment, and it does not imply a second copy in a second place.

None of these is dishonest, but they are technical terms with narrow meanings in ordinary sentences. To fix that, find the sentence in the documentation where the term is defined and quote that sentence. Where no such sentence exists, that absence is the finding.

Published rankings of photo services rarely check any of this, which is why an afternoon of your own testing beats them.

Common questions

Should I strip location data before uploading?

Decide per body of work, not in general. For anything photographed at home, strip the location. For anything where the location is the point, keep it and know it is public.

The key is knowing which of the two you are doing. Your camera and editing software set the default, not you.

Is a checksum overkill for personal photographs?

It costs one command and it turns a vague worry into a yes or a no. You do not need it for every file; you need it for the three test images, so that the round trip test gives a definite answer rather than an impression.

How often should the round trip be repeated?

Once a year, and immediately after any change to the service's storage or pricing tiers, since those are the changes that alter what is retained. Record each run with its date so the results form a series.

What if the export cannot be run without a paid tier?

Then the export is a paid feature and the entry should say so plainly, because that is a material fact about what leaving costs. Test it during a paid month if you have one, and record which tier the observation came from.