Why does AI ignore part of my instruction?

Three causes cover most cases. The request bundles many changes, so the model satisfies the most prominent ones. A word is vague ("nicer", "more premium") and gives the model nothing concrete to change. Or two parts conflict, such as "keep everything the same" next to "change the light".

Placement is a separate weak spot. OpenAI's documentation notes that its image model may have difficulty placing elements precisely in structured or layout-sensitive compositions, so "put the logo exactly in the top left corner" can be followed loosely even when the wording is clear.

What helps: one change per request, concrete words ("warmer, late afternoon light from the left"), and a short list of what must stay. If a part is still ignored after a clear second request, try a new generation that describes the whole scene with that element included from the start.

ImageQuarry is built for this back and forth: you select a version, ask for the change in plain words, and the result appears as a new version beside the old one, so you can compare them side by side and see whether the instruction landed.