AI-Generated Images: Illustration, Never Evidence

An AI image can show an idea. It cannot show a fact. Keep those two jobs apart and you will rarely get into trouble.

A text-to-image tool makes a picture of what a prompt describes, built from patterns in its training. Nothing in that picture happened. No camera was in the room. So the moment an AI image is used to suggest "this is what it looked like," it becomes fabrication, however good the intentions.

When AI images are fine

AI images work well where an editor would otherwise commission a drawing or use a stock graphic:

• Explainers. A stylized picture of water moving through a treatment plant, for a piece on why rates are rising.
• Concepts with no real scene. "A town's budget pulled in four directions." There is nothing to photograph.
• Opinion and essay art. A commentary piece on a road bond can carry an illustrated header, the way a magazine column carries a drawing.
• Promotion. A graphic announcing your podcast episode or a community forum.

In every case the image is plainly an illustration and the article never implies otherwise.

A style choice helps. Illustrations that look like illustrations (flat color, painterly texture, cut paper, line art) are harder to mistake for photos than photorealistic output. If you want readers to understand at a glance that a picture is art, make it look like art.

When they are not

That rules out:

• "What the flood looked like," made because you did not get a photo.
• A realistic image of a named official, candidate or resident in any situation.
• Reconstructions of incidents presented as anything but reconstructions.
• "Enhancing" a real photo with AI fill so that new objects or people appear in it.
• Generated "documents," letterheads or screenshots.

The professional codes line up on this. The SPJ Code of Ethics tells journalists to never publish fabricated content that could be confused with reality and to clearly label illustrations and reenactments. The NPPA Code of Ethics says editing should maintain the integrity of a photo's content and context. A generated picture of a real event fails both tests before you even look at it.

There is also a self-interest reason. The fastest way to destroy a true story is to put one fake image in it. Opponents will ignore your 14 verified records and spend a week on the picture.

Checking images that come to you

The same rule applies in reverse. Realistic AI images circulate constantly, and so do real photos from other years and other states. When a reader sends you a dramatic picture, treat it like any other unverified claim: find who took it, when and where. Ask for the original file. Look for the same image elsewhere with a reverse image search. If you cannot trace it, do not run it as real.

Disclosure that actually works

Disclosure has one job: a reader who sees the image should know how it was made. That means the label goes with the picture. A policy page nobody reads does not count.

Put it in two places:

1. The caption or credit line, every time.
2. The image itself, in small text, when the picture might be shared on its own. Social posts, link previews and screenshots strip away your caption. A small "AI illustration" in a corner travels with the pixels.

Disclose AI editing on real photos too, when it changes content. Removing a power line with generative fill changes what the scene showed. Ordinary adjustments like brightness, contrast and cropping do not need a label as long as they keep the image truthful.

Who owns an AI image?

In January 2025 the U.S. Copyright Office published Part 2 of its report on copyright and artificial intelligence. Its conclusions: copyright does not extend to purely AI-generated material, and prompts alone, with today's widely available tools, do not give a person enough control over the output to count as its author. Human contributions can still be protected, such as your own arrangement of elements or substantial hand edits.

So your raw AI images likely give you thin or no copyright protection, so others may be able to reuse them. And check the terms of whatever service made the image, since those terms govern what you agreed to.

Making the image: ImageStudio's three sizes

ImageStudio makes AI images at three sizes. Pick the size before you write the prompt, because the shape changes what the picture can hold.

Write prompts that make the illustration obvious: name a style ("flat vector illustration," "watercolor," "paper cutout"), and keep real names, real logos and real places out. "Watercolor illustration of a small-town water tower with a rising arrow, muted blues" works. A prompt that names your actual mayor does not, for every reason in this lesson.

Generated text inside images is often garbled. Add headlines and labels afterward in a design tool, where you control the spelling.