The short answer
- Say how the image was made with IPTC Digital Source Type, using one term from IPTC's controlled vocabulary.
- Say which AI was used with the four AI fields added in IPTC Photo Metadata 2025.1: AI System Used, AI System Version Used, AI Prompt Information and AI Prompt Writer Name.
- Say whether it may train AI with the PLUS Data Mining property.
All of these live in the file's XMP metadata, so they travel with the image to anyone who receives the original. Google and Meta read Digital Source Type; other platforms, such as LinkedIn and TikTok, label from C2PA Content Credentials instead.
Which term to use
IPTC's guidance for the common cases. Each term is the last part of an address that starts with http://cv.iptc.org/newscodes/digitalsourcetype/. Term names are IPTC's (credit below the full vocabulary).
| Your image | Term | IPTC name |
|---|---|---|
| Generated entirely by an AI model from a prompt | trainedAlgorithmicMedia | Created using Generative AI |
| A photo edited with generative AI (inpainting, outpainting, Generative Fill) | compositeWithTrainedAlgorithmicMedia | Edited using Generative AI |
| A composite where at least one element is generative AI | compositeSynthetic | Composite including generative AI elements |
| Enhanced by an algorithm that does not change the content (noise reduction, sharpening) | algorithmicallyEnhanced | Algorithmically-altered media |
| A straight photograph from a camera | digitalCapture | Digital capture sampled from real life |
| A photograph retouched by a person with non-generative tools | humanEdits | Human-edited media |
| A smartphone shot merged from several frames (HDR, night mode) | computationalCapture | Multi-frame computational capture sampled from real life |
Scroll the table sideways for the remaining columns
IPTC's guidance also accepts compositeSynthetic for AI-generated images, and gives no
specific rule for AI upscaling or sky replacement. A sensible reading is that anything a generative
model added or changed is compositeWithTrainedAlgorithmicMedia, and a sky taken from
another real photo is compositeCapture.
The full vocabulary
IPTC's Digital Source Type vocabulary, current release 23 October 2024. Three older terms are retired:
minorHumanEdits (use humanEdits), digitalArt (use
digitalCreation) and softwareImage.
| Term | IPTC name | Meaning |
|---|---|---|
digitalCapture | Digital capture sampled from real life | Captured from real life with a digital camera or recorder. |
computationalCapture | Multi-frame computational capture sampled from real life | Several captured frames merged automatically, by signal processing or non-generative AI. |
negativeFilm | Digitised from a transparent negative | Scanned from a film negative. |
positiveFilm | Digitised from a transparent positive | Scanned from a slide or transparency. |
print | Digitised from a non-transparent medium | Scanned from a print or another opaque original. |
humanEdits | Human-edited media | Changed by people with non-generative tools. |
compositeWithTrainedAlgorithmicMedia | Edited using Generative AI | Changed with a generative AI model, for example by inpainting or outpainting. |
algorithmicallyEnhanced | Algorithmically-altered media | Corrected or enhanced by an algorithm a person started, without changing the main content. |
digitalCreation | Digital creation | Made by a person with non-generative digital tools. |
dataDrivenMedia | Data-driven media | A visual representation of data. |
trainedAlgorithmicMedia | Created using Generative AI | Created by an AI model trained on captured content. |
algorithmicMedia | Pure algorithmic media | Created by an algorithm without training data, such as a fractal. |
screenCapture | Screen capture | A capture of a computer or phone screen. |
virtualRecording | Virtual event recording | A recording of a virtual event made from generative or captured elements. |
composite | Composite of elements | A mix of elements, any of which may or may not be generative AI. |
compositeCapture | Composite of captured elements | A mix of elements that are all real-life captures. |
compositeSynthetic | Composite including generative AI elements | A mix of elements, at least one of them generative AI. |
Scroll the table sideways for the remaining columns
Terms and names: IPTC Digital Source Type NewsCodes, © IPTC, International Press Telecommunications Council, iptc.org, licensed under CC BY 4.0. Meanings paraphrased.
The AI fields in IPTC 2025.1
Added to the IPTC Extension schema in 2025. ExifTool writes them from version 13.40 (a development release); the first production release with them is 13.44.
| Field | ExifTool tag | What to put in it |
|---|---|---|
| AI System Used | XMP-iptcExt:AISystemUsed | The AI tool or model, such as the image generator you used. |
| AI System Version Used | XMP-iptcExt:AISystemVersionUsed | Its version, if known. |
| AI Prompt Information | XMP-iptcExt:AIPromptInformation | The prompt or prompts, including negative prompts. |
| AI Prompt Writer Name | XMP-iptcExt:AIPromptWriterName | Who wrote the prompt. IPTC notes this person is not necessarily the creator. |
Scroll the table sideways for the remaining columns
Opting out of AI training
The PLUS Data Mining property says whether an image may be used for data mining, including AI
training. Its values (stored with the prefix http://ns.useplus.org/ldf/vocab/):
| Value | Code |
|---|---|
| Unspecified (no prohibition defined) | DMI-UNSPECIFIED |
| Allowed | DMI-ALLOWED |
| Prohibited for AI/ML training | DMI-PROHIBITED-AIMLTRAINING |
| Prohibited for Generative AI/ML training | DMI-PROHIBITED-GENAIMLTRAINING |
| Prohibited except for search engine indexing | DMI-PROHIBITED-EXCEPTSEARCHENGINEINDEXING |
| Prohibited | DMI-PROHIBITED |
| Prohibited, see Other Constraints | DMI-PROHIBITED-SEECONSTRAINT |
| Prohibited, see Embedded Encoded Rights Expression | DMI-PROHIBITED-SEEEMBEDDEDRIGHTSEXPR |
| Prohibited, see Linked Encoded Rights Expression | DMI-PROHIBITED-SEELINKEDRIGHTSEXPR |
Scroll the table sideways for the remaining columns
With "see Other Constraints", write your terms in the Other Constraints field. The standard notes that regional law may override this property, and that leaving it empty does not grant permission.
Who reads these fields
- Google recommends Digital Source Type to creators and may show how an image was made in "About this image". Merchant Center requires AI-generated product images to keep
trainedAlgorithmicMediaand not strip it, and Google Photos shows AI edits from IPTC metadata. - Meta said in February 2024 that it detects the AI signals in the IPTC and C2PA standards to label images on Facebook, Instagram and Threads.
- The EU AI Act (Article 50, from 2 August 2026) requires providers of generative AI systems to mark their output in a machine-readable way, except where a tool only assists standard editing, and requires those who use AI to create deepfakes to disclose them. Digital Source Type is one machine-readable way to carry such a mark. This is not legal advice.
IPTC metadata and C2PA
C2PA Content Credentials record how a file was made in a signed manifest, so tampering is detectable, and they use the same IPTC Digital Source Type terms. IPTC metadata is plain and editable, which makes it easy to add, correct and read in any metadata tool. They complement each other, and both can be lost when a platform rebuilds an uploaded image, so keep your originals with their metadata.
How to do it in Parergon
- Select the photo and open the IPTC Extension section of the editor.
- Pick a term from the Digital Source Type list. It offers IPTC's full vocabulary by name, labels the three retired terms as retired, and stores the official address.
- Fill in AI System Used, AI System Version Used, AI Prompt Information and AI Prompt Writer Name as needed, and pick a Data Mining value from the PLUS list.
- Save. To mark a whole shoot the same way, put these fields in a template or metadata profile and apply it to the selection (Pro).
Editing these fields one file at a time counts toward the free tier's five edits a day. For every field Parergon writes, see the field reference.
With ExifTool
Tested with ExifTool 13.55. Digital Source Type takes the full address; Data Mining takes the value's name.
exiftool -XMP-iptcExt:DigitalSourceType=http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia \
"-XMP-iptcExt:AISystemUsed=Image generator name" "-XMP-iptcExt:AISystemVersionUsed=2" \
"-XMP-iptcExt:AIPromptInformation=A lighthouse at dusk" "-XMP-iptcExt:AIPromptWriterName=Jane Doe" image.jpg
exiftool "-XMP-plus:DataMining=Prohibited for Generative AI/ML training" photo.jpg
exiftool -G1 -XMP-iptcExt:DigitalSourceType "-XMP-iptcExt:AI*" -XMP-plus:DataMining image.jpg