Everyone uses `--ar` and thinks they're composition pros. They're not. The real control is in region-specific prompting, and most tutorials overcomplicate it.
Here's the stripped-down workflow that actually works for me.
* Start with `/prefer remix` to keep the prompt editor open.
* Use `/show` on an image you like, then `Vary (Region)`.
* The key: your new prompt **must** describe *only* the region you're changing. Midjourney doesn't re-process the whole scene, it blends your new description into that area.
* Example: You have a portrait, but want a specific hat on the subject.
* Bad: "a person in a forest wearing a red beret" (This will try to remake the whole person/forest).
* Good: "a detailed red beret" (This targets just the head area).
Tried this on a product shot last week. Original: "a sleek black coffee maker on a marble counter." Selected the steam region, prompted "a stylized bear shape formed from steam." It worked on the third try, because the prompt was local.
Biggest pitfall? Being too descriptive. Keep the region prompt focused and simple. It's a surgical edit, not a scene rewrite.