Sidecar Files and Embedded Metadata: How AI Photo Tags Travel Between Apps

Three places where photo descriptions can live

A digital photograph is more than its visible pixels. It may also contain embedded information such as the date and time of capture, camera details, orientation and location. Additional descriptions can be stored separately in a sidecar file, while an application may maintain its own catalogue of captions, people tags and classifications. These are three distinct storage locations, even when the application presents them together on one screen.

This distinction matters when moving a collection between services. Downloading an image file retrieves the image and whatever metadata is embedded in it. It does not automatically retrieve every description associated with that image in the original application. A caption visible beside a photograph may be a catalogue entry, an embedded field or a separate record. Its appearance alone does not identify its storage location.

• Inside the image: embedded metadata.
• Beside the image: a sidecar file.
• Inside the service: an application-specific catalogue or database.

What an XMP sidecar file contains

XMP stands for Extensible Metadata Platform. It provides a structured way to describe digital files using named metadata fields. An XMP sidecar is a separate file containing descriptive information associated with an image. It can hold details such as a title, description, keywords, copyright information and other supported metadata. The image itself remains a separate file, so the description can be maintained without rewriting the photograph.

For example, a photograph named garden-1984.jpg might have an associated garden-1984.xmp file. The two files belong together as an archive unit. The sidecar can contain descriptive information while the JPEG retains its original image data. However, the exact fields depend on the application and its export settings. A sidecar should not be assumed to contain every field visible in the editing interface.

• An XMP file is metadata, not another copy of the photograph.
• It can be read or edited by compatible metadata software.
• Its usefulness depends on supported fields and continued association with the image.

Why AI photo tags may remain inside an app

AI photo tagging commonly involves identifying subjects, suggesting captions or grouping images by apparent content. These descriptions may be stored in a service's internal catalogue rather than written into the photograph. The application can then display a person, place or subject tag without changing the original image file. This arrangement allows the service to manage its own search and classification system.

When exporting, distinguish between an image download and a metadata export. The first may contain only the photograph. The second may provide a separate description file or structured catalogue, depending on the available options. Some services offer both, while others provide limited export choices. A successful image download therefore does not establish that the AI-generated descriptions have accompanied it.

A practical method for exporting tags

Start by identifying which information must survive outside the original application. For a family collection, this might include the original capture date, a descriptive caption, a place name and the names of identified people. Separate information that describes what is visible from information that records who supplied or confirmed a description. An automatically suggested name should not be treated as a confirmed identification.

Next, inspect the application's export choices. Look for options referring to metadata, descriptions, keywords, XMP files or original image information. If a separate catalogue or structured data export is offered, retain it as well. Keep the original image files unchanged and place associated sidecars in the same directory. Record the application's name and export date in a short accompanying note.

• Identify the fields that matter before exporting.
• Choose original-quality image files where available.
• Include separate metadata or catalogue exports when offered.
• Keep each sidecar with its corresponding image.
• Retain an untouched copy of the original collection.

Opening the archive outside the original application

A portable archive should be understandable without access to the account that created it. Open a selection of photographs in ordinary image software and inspect their embedded information. Open any XMP sidecars in a compatible metadata reader or text editor. Confirm that the image and its descriptive information can still be associated. If a field appears in the original application but nowhere in the exported files, record that limitation in the archive inventory.

Consider a simple worked example. A collection contains 240 photographs, each with a capture date and an AI-generated subject label. The export includes 240 JPEG files and 180 sidecars. The missing 60 sidecars require attention: some photographs may have no additional metadata, or the application may have exported only selected descriptions. Record the difference and preserve any separate catalogue information that remains available.

Keep a record of what each file preserves

The archive should describe the relationship between the photograph and its accompanying information. A short inventory can identify the image format, the metadata format, the original application and the fields expected in each location. It should also state whether the date represents capture, editing or export. This prevents a later keeper from treating several different dates as interchangeable.

A practical record can distinguish information retained in the image, information supplied by a sidecar and information that exists only in a separate application export. Use clear labels such as present, absent and application-dependent. When a description has been manually corrected, retain that distinction in the record. The next step is to select one photograph with a sidecar and confirm that both files are stored together in the archive.

• Image file: original photograph and embedded metadata.
• Sidecar file: separately stored descriptive fields.
• Catalogue export: additional application records, if provided.
• Inventory note: format, source application and export date.

Questions readers ask