You tried AI staging once, got back a room with the couch still sitting in the middle of it, and decided the tool was not ready. It was a wording problem rather than a capability one, and the fix is not a longer list of objects — it is one word about where.
An Austin agent ran a dated living room from his own listing through Google's Nano Banana image model inside Gemini, and published every prompt he used, including the ones that did not work. In order:
1. "Remove all of the furniture in this picture." It cleared the furniture in the background and left the pieces in the middle of the room.
2. "Remove the couch and the carpet and the turntable and TV table." Still incomplete.
3. "Remove the furniture in the middle of this room." That did it. The room came back clean.
4. To stage it: "Add a modern set of furniture in the middle of the room." He describes a contemporary sofa, coffee table, area rug, and accent pieces arriving with the room's own perspective, proportions, and shadows.
His diagnosis of the first failure is that if you are not specific enough about location, the model makes its own assumptions about what to remove. He also says he prefers this two-step order — empty first, stage second — because the model is then not trying to blend new furniture over the outlines of the old pieces. He puts the cost at zero through Gemini, against the $25 to $75 per photo he quotes for virtual staging services.
“The first attempt removed the furniture in the background but left the pieces in the middle of the room.”
AI staging has a bad reputation with agents for a reason that has nothing to do with the models. People try it once on a real listing, get a half-emptied room, and conclude the technology is a year away. A general instruction lets the model decide what counts as furniture, and it guesses.
Naming the location instead of the objects is the whole fix, and knowing that three attempts is normal is what keeps you going past attempt one. What it does not fix is disclosure. An altered photo is still an altered photo, and that is a rule question, not a taste question.