The selfie is usually fine. The kitchen is not. I keep the face, hair, and pose, and I replace only the background with a plain wall or a soft, out-of-focus interior. I do not request a younger face or a different jaw. A bio photo has to survive a meeting in person.If I also need a darker shirt, that is a second pass. Wardrobe and room at the same time is how shoulders drift.
I do not run a studio, and I do not "explore generative art" on weeknights for fun. I have a job that produces leftover pictures: phone packshots, a dim photo from an event, a selfie that has to stand in for a bio, a screenshot that needs a wider card, a sketch that has to be shown to someone who does not want to look at a sketch. This is a list of seven jobs I now finish in an AI image editor with a still and a short instruction, instead of booking a reshoot or opening a file I will never complete.
Each item is a constraint I actually type, not a feature tour. If a result looks expensive and the subject has changed, I throw it away and do the job again, or I leave the original alone.
A short rule I apply to all seven
I write the lock before the change, I do one change per generate, and I compare the output to the source at the same size. The editor is prompt-based, it accepts everyday still formats, and it can hold a character or a product across a scene when I attach a clean reference. That is enough for a work week. The point of the list is not to make more pictures. It is to finish the seven I was already going to need, without pretending I had a studio day I did not have.