isGenAI.com

Glossary

AI-generated image: what the term means and how it is recorded

An AI-generated image is a picture whose pixels were produced by a generative model such as a diffusion or transformer model from a text prompt, a reference image or both, as opposed to a photograph captured by a sensor or an illustration drawn by hand.

Definitions and isGenAI behaviour as of October 2026. Updated .

See the term in a real reportFrequently asked questions
Code-drawn illustration for AI-generated image

Illustration · PNG file

  1. Signature + IHDR
  2. iTXt XML:com.adobe.xmprecord
  3. tEXt parameters
  4. tEXt prompt / workflow
  5. tEXt invokeai_metadata
  6. eXIf
  7. caBX C2PA manifest
  8. IDAT pixels
  9. IEND

AI source type declared

The file labels itself as AI-made.

IPTC DigitalSourceType such as trainedAlgorithmicMedia is read from XMP and shown with its exact value.

Check an image →
AI source type declared: where this record sits inside a PNG file and what the isGenAI report says about it. The picture is a code-drawn illustration.

How generation is recorded

Generators record the fact in different ways: signed C2PA manifests (OpenAI, Adobe Firefly, Black Forest Labs, newer Google models), IPTC DigitalSourceType declarations in XMP (Midjourney, Meta AI, Microsoft), saved prompts and workflows in PNG text chunks (AUTOMATIC1111, ComfyUI, InvokeAI) and invisible watermarks (Google, OpenAI, Meta). Many hosted tools record nothing we could find. The generator pages list what each vendor documents.

Edited photographs with AI fill are a separate category, often declared as compositeWithTrainedAlgorithmicMedia.

How isGenAI reports it

isGenAI answers “is this image AI-generated?” from the file: it reveals the signed record, the declared source type, the saved prompt, the AI statement and any matching copy, each labelled with its source and strength. It does not estimate from the picture, so a file with no records is reported as no evidence found with a plain statement that this does not prove human authorship.

Why the distinction matters

Article 50 of the EU AI Act attaches duties to AI-generated content: providers must mark it machine-readably and deployers must disclose deepfakes and certain public-interest text. Being able to show the file record is how both sides demonstrate compliance.

Hybrid workflows blur the line further. A photograph can be extended with generative fill, a render can be painted over by hand, and a sketch can be finished by a model. The records, where they exist, describe those steps more precisely than any single label.

How to use this term when checking a file

  1. Upload the original file to the AI image detector or the tool named on the page.
  2. Find the finding that uses this term under “How we checked” and read the source field it names.
  3. Weigh it by kind: a signed record, a declared field, a saved prompt and a matching copy are different strengths of evidence.
  4. Remember that absence of a record is not evidence of a human author.

See the term in a real report

Frequently asked questions

How can I tell if an image is AI-generated?

Check the original file for a C2PA record, an IPTC source type, a saved prompt or an AI statement with isGenAI, trace the source, and treat a clean result as unknown rather than human-made.

Do all AI-generated images carry evidence?

No. Many hosted tools write nothing, and platforms strip what is written. Absence of records is not proof of a photograph.

Does isGenAI guess from the pixels?

No. It reads file records and matches copies, and says when it finds nothing.

Is an AI-edited photo an AI-generated image?

Partly. IPTC uses compositeWithTrainedAlgorithmicMedia for mixes, and C2PA manifests can record individual edit actions.

Sources

Definitions follow the published standards linked above; the isGenAI behaviour described is the deployed checker as of October 2026.