Make an image, then change it

A pure image model takes a prompt and nothing else. It cannot look at the picture you already have. An editing model takes your picture plus an instruction, and changes what you named.

This catches people out because the names do not say so. The same family often has both, at two addresses. Attach a picture and the studio moves to the one that reads it; you do not have to know which is which.

Edit when most of the frame is right and one thing is wrong: the jacket is the wrong colour, there is a logo you cannot use, the background is cluttered. The rest of the picture survives, which is the entire point.

Start again when the composition is wrong. Asking an editor to change the framing is asking it to redraw everything, and then you have paid for an edit and received a worse generation.

Name one thing. Everything you describe is something the model may decide to redraw, so a full description of the scene is an instruction to touch all of it.

Editing models take one or more source images, and the maximum comes from the maker. It is not the same everywhere: one model here stops at four, another accepts sixteen. The studio shows the real number and will not let you attach one more than the model takes.

That sounds like a small thing. It is the difference between a refusal you see before you pay and a refusal you get back from the provider after the credits are gone.