What can I do when an AI image comes out wrong?

Sort the problem first. A wrong detail (a color, a prop, the light) is an edit. A wrong idea (the subject, the camera angle, the arrangement of the scene) is a new generation, because each edit starts from an image whose basic structure you already dislike.

When you edit, ask for one change per turn and say what must stay: "Make the background plain gray. Keep the person, the crop and the light." Several changes in one request make it harder to see which instruction caused which result.

Some problems are known weak spots of current image models rather than prompt mistakes. OpenAI's documentation lists precise text, consistency of recurring characters or brand elements, and exact placement in layout-sensitive compositions as limitations. For exact text or logos, add them afterwards in a design tool.

In ImageQuarry, each change you ask for becomes a new version next to the earlier ones, so a fix that goes the wrong way never costs you the image you had. You can return to an earlier version and branch from it instead.