A photograph begins with a camera recording light from a scene. A generated image can be constructed without that scene existing. An edited photograph starts with captured material but may change it slightly or substantially. Modern workflows often combine these categories, so the useful question is what changed and which parts still support a factual claim.
The amount of visual change is not the only issue. A small alteration to a date on a notice can matter more than a large colour adjustment to a decorative image.
Conventional editing can change meaning
Cropping, exposure adjustment and colour correction are familiar photographic edits. Their effects depend on context. Cropping an empty border may be unimportant; cropping a person out of a scene can change how an event is interpreted.
Compositing can combine material from several photographs without using a generative model. Removing an object or replacing a background can also alter the apparent facts. Consequently, “not made by AI” should never be treated as equivalent to “unaltered evidence”.
The intended use matters. An illustration for a fictional story and an image presented as documentation of a real incident create different expectations for viewers.
Generative editing introduces invented material
Generative tools can create content within a selected area or add material beyond an original frame. Adobe's documentation for Generative Fill is one example of this type of editing capability. A photograph can therefore remain recognisably photographic while containing newly generated details.
Consider a hypothetical photograph of a garden extended to fit a wide banner. The original centre may be camera-captured, while the additional plants at either edge are synthetic. Calling the entire image a camera photograph without mentioning the extension can leave readers with the wrong understanding of its history.
This is also why a model's convincing output is not evidence of hidden details. As explained in how AI image generation works, the system can construct plausible content without recovering an observed fact.
Use a description that matches the change
“Photograph with colour adjustments” communicates something different from “photograph with an AI-generated background”. “AI-generated illustration” is more appropriate for a wholly synthetic scene. These are examples of descriptive labels, not universal legal formulas.
Choose wording based on what would affect a reasonable viewer's interpretation. If a background was replaced, say that. If the image is a composite, explain the combination where it matters. Avoid a vague label such as “enhanced” when the change introduced people, objects or events absent from the source.
Keep the original and the editing history
For your own work, retain the source file separately from the edited export. Record significant transformations so that you can explain them later. Where supported, preserve Content Credentials and other provenance information rather than relying entirely on memory or a filename.
A record of edits can help an audience understand the process. It does not establish that every statement accompanying the image is true. A genuine photograph can still be posted with the wrong date or an invented caption.
Assess the claim as well as the pixels
Before sharing a disputed image, identify what the caption asks you to believe. Is it merely showing a design concept, or claiming that a particular event happened? Check the source and context accordingly.
When the available evidence cannot distinguish capture, editing and generation, describe that uncertainty. The guide to detecting AI-generated images explains why an automated score or one unusual detail is not enough to settle every case.
Related reading
How AI Image Generation Works: A Beginner’s Guide
