IPTC DigitalSourceType: The Field That Tells Platforms an Image Is AI

One field, a fixed list of answers

DigitalSourceType is part of the IPTC Photo Metadata Standard, the vocabulary newsrooms, stock agencies and photo software use to describe images. It answers one question: how was this content made? The answer is not free text but a term from the IPTC “digital source type” controlled vocabulary, written as a URI such as http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia.

In image files the field is stored in XMP as Iptc4xmpExt:DigitalSourceType. C2PA Content Credentials reuse the same vocabulary inside their signed actions, so the same terms appear in both places.

The values you will meet

The vocabulary has grown over the years. The terms most relevant to AI labels are:

• trainedAlgorithmicMedia — created by a generative model trained on sample content.
• compositeWithTrainedAlgorithmicMedia — a composite that includes generative AI elements, for example a photo edited with Generative Fill.
• algorithmicMedia — created purely by an algorithm without training data, such as procedural or fractal art.
• compositeSynthetic — a composite that includes synthetic elements.
• algorithmicallyEnhanced — minor augmentation or correction by an algorithm.
• digitalCapture — captured by a digital camera or scanner from a real scene.
• computationalCapture — a capture processed by algorithms at the time it was taken, as many smartphones do.
• digitalArt and screenCapture — artwork made by a person with software, and screenshots.

Who writes it

Several AI image services write trainedAlgorithmicMedia into their output. Editors that offer generative tools can record a composite value after an AI edit. Stock libraries and news agencies fill it in during ingestion, and some marketplaces ask sellers to mark AI images this way. The field can also be copied forward silently: once set, it survives every app that preserves XMP.

Who reads it

Meta has said it uses IPTC metadata alongside C2PA when deciding whether to show “AI info” on Facebook, Instagram and Threads. Other platforms, search engines and shopping services have announced similar use of the field for AI disclosure. That is why a lightly retouched real photo carrying compositeWithTrainedAlgorithmicMedia can end up labelled exactly like a fully generated image.

Unsigned, so easy to change

Unlike a C2PA manifest, plain IPTC metadata carries no signature. Any metadata editor can add, change or remove DigitalSourceType, and many apps drop it when re-saving. Treat the field as a declaration, not as proof — whether it says “AI” or “camera”.

Finding and removing it

Lay the image in AI Label Remover. In “AI traces” the field appears as “XMP · Iptc4xmpExt › DigitalSourceType” with its value; the same term inside a C2PA manifest is listed under C2PA. Pressing “Wash all images” removes the whole XMP packet and C2PA store, which takes the field with it, along with EXIF and other identifying data. The AI metadata remover page covers the other AI fields too.

If you need to remove only this one field and keep the rest of your XMP, a command-line tool can do it: exiftool -XMP-iptcExt:DigitalSourceType= photo.jpg. Remember that a matching value may also sit in a C2PA manifest.

When the field should stay

The field exists so viewers know how an image was made. If an image really is generated or substantially altered by AI, keep the declaration — or disclose it when you post. Removing it is appropriate when it misdescribes your work or when you are cleaning private data from a file.