Image tools generate from descriptions, edit what exists, and replace parts of a picture. Each of those raises a different rights question.
Do now: Generate three visual directions for something you actually need, and write down what each one implies about reality.
Ask what the image claims, not just how it looks.Visual work is where AI most obviously saves money and most easily creates legal and ethical exposure. The technique takes an hour to learn. The habits around consent, likeness and honesty are what keep the work usable.
Three things these tools actually do
Generation makes a new picture from a text description. Editing changes an existing picture — extending it, adjusting light, removing an object. Replacement swaps part of a picture for something else, which is the most powerful and the most dangerous, because the result still looks like a photograph of a real moment.
Good briefs describe subject, setting, composition, light, mood and format. Vague briefs produce the average of everything the tool has seen, which is why so much AI imagery looks like the same glossy nowhere. Specificity is what makes output yours rather than generic.
Iterate in one dimension at a time. Change the light, keep everything else. Change the composition, keep the light. Changing five things at once tells you nothing about which change helped.
A generic brief produces a generic image. The tool is averaging; your job is to stop it.
Rights, likeness and the honest question
Generating an image does not automatically give you ownership of it, and rules differ by country. Separately, tools have their own licence terms about commercial use — free tiers often restrict it. Check both before an image goes on a product, a poster or an advertisement.
Likeness is its own category. Real people's faces, and styles closely associated with a living artist, carry risks that have nothing to do with whether the tool let you make them. Consent for a photograph is not consent to alter it, place it in a different setting, or make the person appear to do something.
Then the honest question: what does this image claim? A clearly illustrative header claims nothing. A photorealistic picture of your premises, your staff or your product claims those things exist and look like that. Using a generated image where a customer would reasonably assume a photograph is a decision about honesty, not about design.
Keep a rights log, from the first image
For every asset record: what tool made it, what prompt or reference you used, whether any input was someone else's work, what the licence permits, and where it is published. This takes a minute per image and becomes impossible to reconstruct six months later when a question arrives.
The log also protects you commercially. A client asking whether they may reuse an asset, a platform asking whether content was synthetically generated, a partner asking whether you cleared a likeness — all of these are quick answers if you logged, and expensive investigations if you did not.
The rights log is not paperwork. It is the difference between answering a question in a minute and losing a week.