“AI image” can describe several different workflows. A detector may not tell which workflow produced a file, especially after multiple edits.
Four common workflows
- Fully generated image: a model creates the image from a prompt or other input, without a camera photograph as its base.
- Retouched photograph: a camera image is adjusted with color correction, cleanup, cropping, or compositing. Retouching can change the appearance while leaving most photographed content intact.
- Composite photograph: elements from multiple sources are combined. Some parts may come from photographs, while others may be generated or painted.
- AI-assisted edit: an editing tool uses a model to remove an object, extend a background, replace details, or alter part of an existing photo.
The categories can overlap. A photograph can be retouched and have one generated region. A classifier trained to separate generated images from camera images may return a strong signal for an edited photo without identifying what changed.
What a detector can tell you
Motchilllz’s AI Analyzer returns class scores for the uploaded image. It does not locate generated regions or reconstruct an edit history. Its result is probabilistic, and compression or later edits can affect it.
EXIF software fields can provide a name recorded in metadata, but they do not list every operation and can be edited or removed. Provenance records, when present and valid, can describe some creation and editing steps. They also do not decide whether the depicted scene is truthful.
When the distinction matters, ask for the original file, project history, or an explanation from the creator. The image alone may not preserve enough information to classify the full workflow.