Can ChatGPT Do Virtual Staging? We Tested It on the Same Room

ChatGPT can stage a room, but in our tests it didn’t stage your room. We ran one empty listing photo through three OpenAI image models, most recently GPT Image 2 on 15 October 2026. All three returned an attractive furnished room in a different shape from the original, with the windows redrawn. Three purpose-built image editors kept every window, the doorway and the photo’s exact proportions. For a photo that will represent the property on the MLS, that difference is the whole job, because MLS rules ban changing the building.

“Can ChatGPT do virtual staging?” turns up in Google’s People Also Ask box under nearly every virtual staging search: the head term, the software comparisons, the pricing queries. It’s the most repeated question in the category, and the answers are uniformly vague.

So we tested it. One empty room, one prompt, six models, outputs compared at full size.

What’s changed since September (checked 15 October 2026). OpenAI has moved on twice: GPT Image 2 arrived on 21 April 2026, and ChatGPT Images 2.5 on 8 September 2026, which OpenAI says is “better at editing only what you’ve asked for.” Both newer API models accept custom output sizes, so our September line that OpenAI’s endpoint “only emits fixed sizes” was out of date for them. We re-ran the room through GPT Image 2 in a chat, the way an agent would upload a photo. It came back as a square, with the room squeezed to fit. The verdict stands; the reason has shifted from “can’t” to “didn’t”.

The test

The room is a generated photorealistic empty living room. It’s synthetic on purpose, so it’s licence-clean and reproducible and the geometry is unambiguous: two rectangular windows on the left wall, a radiator beneath them, a doorway on the right, oak floor, crown moulding. It measures 1152 × 896 pixels, an aspect ratio of 1.286.

Every model got the identical prompt: stage it as a warm modern living room, add sofa, armchair, coffee table, rug, floor lamp, plant and wall art, and don’t change the walls, ceiling, windows, window frames, doorway, radiator, baseboards, floor material or camera angle.

That last clause isn’t decoration. Every MLS rule on virtual staging says a version of the same thing: you may add furniture, you may not change the building.

The results

Model	Cost per image	Time	Aspect kept	Architecture kept
Gemini 3.1 Flash Lite	$0.0339	6.3–7.6s	✅ 1.286	✅
Gemini 2.5 Flash Image	$0.0391	~9s	✅ 1.286	✅
Gemini 3.1 Flash Image	$0.0678	11.6s	✅	✅
GPT Image 2 (via chat, Oct 2026)	$0.237	110s	❌ forced 1.000	❌
GPT-image-1 (Sept 2026)	—	—	❌ forced 1.500	❌
GPT-image-1-mini (Sept 2026)	—	—	❌ forced 1.500	❌

Costs are the billed amounts returned by the API, not list-price estimates. One run per model, default settings, one synthetic room.

What the OpenAI models got wrong

They didn’t return your photo’s shape. gpt-image-1 and gpt-image-1-mini only emit fixed sizes (1024×1024, 1536×1024, 1024×1536), so our 1.286:1 room came back at 1.500:1. GPT Image 2, given the photo in a chat, came back as a 1,024-pixel square. Before you judge a single piece of furniture, the room has been re-proportioned and the framing changed.

They redrew the windows. On close crop, gpt-image-1 returned windows that are shorter, wider and differently spaced than the originals, with the wall-to-ceiling junction moved. gpt-image-1-mini was worse: the two separate windows widened into what reads as a single window unit, and the radiator vanished completely. GPT Image 2 kept two windows and the radiator, but squeezed the whole room horizontally to fit the square, so both windows came back narrower than they are.

All three outputs are attractive rooms. Shown on their own, most people would call them excellent staging. Put them next to the original and they’re a different property.

Can you fix the shape now?

Partly, and only through the API. OpenAI’s own guide says GPT Image 2 “supports any resolution passed in the size parameter” within limits, and the image generation guide says the 2.5 models take custom sizes whose width and height “must be multiples of 16”. Our 1152 × 896 photo fits those rules, so a developer can now ask for the right shape. We haven’t tested that route, or Images 2.5 inside the ChatGPT app. Asking for the right frame also doesn’t stop a model redrawing the windows inside it, which is the failure that breaks MLS rules.

What the purpose-built models got right

All three Gemini image models returned the room with the same two windows, the same mullion grid, the same spacing, the same sill and trim, the same doorway, the same floor, and the same aspect ratio. The furniture is added; the building is untouched. The October re-run of Gemini 3.1 Flash Lite did the same, at the same $0.0339.

One caveat: in one September run, a Gemini model placed a sideboard where the radiator was, hiding it. Real stagers put furniture in front of radiators all the time, so it’s defensible, but it changes what the buyer can see.

What ChatGPT and Gemini say about this

We asked both “Can ChatGPT do virtual staging for a real estate listing photo?” on 15 October 2026 (one run each). ChatGPT answered “Yes!” and added, fairly, that “AI can unintentionally alter windows, doorways, flooring, or room proportions.” Gemini said “yes, but with major limitations”, which matches our result, but two of its details are wrong:

• It said ChatGPT’s image generator is “DALL-E”. It isn’t: ChatGPT runs ChatGPT Images 2.5, launched on 8 September 2026.
• It said ChatGPT “outputs images at fixed aspect ratios”, citing this page’s September version. That’s true of the older models we tested, but OpenAI’s newer API models take custom sizes. The shape problem is now about what you get in practice, not a hard limit.

Why this is a compliance problem, not a taste problem

If this were only about quality, the answer would be “use a better tool”. It isn’t.

NorthstarMLS’s guidelines prohibit changes to “walls, floor coverings, doors/windows, roofing, siding, ceilings, fireplaces, driveways, site grading, or other permanent features.” Canopy MLS prohibits “changing room dimensions” and adding features that don’t exist. CRMLS, enforcing California’s AB 723 since 1 January 2026, bars modifying “any real part of a property such as wall color, flooring, cabinets, wall dimensions, landscaping”.

An image with re-proportioned windows breaks those rules on its face, and the agent has no idea, because they disclosed the thing they did (added furniture) and not the thing the model did (moved the window). Disclosure doesn’t cure it. You can’t label your way out of showing a property that doesn’t exist.

That’s the real answer to “can ChatGPT do virtual staging”. Not the quality is a bit worse. It’s this: the failure is invisible to you and puts you on the wrong side of an MLS rule.

When ChatGPT is useful here

None of this makes ChatGPT useless for listing work. It’s the wrong tool for this one job. Things it does well:

• Writing the listing copy around the staged photo. It’s good at this, and we have twelve prompts for it.
• Choosing a furniture style for a property. Describe the house and ask what would suit the likely buyer; it’s a reasonable design consultant.
• Concept renders that aren’t listing photos. Renovation ideas for a seller conversation, mood boards, “what could this room become”. Nobody treats those as a record of the property.
• Drafting your disclosure language.

The line is whether the output will be presented as a photograph of the property. Inside that line, use a model that preserves geometry. Outside it, ChatGPT is fine.

What to use instead

Purpose-built staging tools use image-editing models that are told to keep the geometry, and they return your photo at its own dimensions. Paid options include VirtualStagingAI and ApplyDesign; free tiers include Remodel AI (three photos with a free account). We compare them in the best virtual staging software and what free actually buys you. Reallyo’s free virtual staging tool is switched off for now.

Whatever you use, do the check that took us five minutes: open the original and the staged version at full size, side by side, and look at the windows. It’s the fastest way to catch a model that has quietly redecorated the architecture.

FAQ

Can ChatGPT do virtual staging?

It can generate a staged room from your photo, but in our tests it didn’t preserve the room. The two older OpenAI models returned fixed shapes and altered the windows; GPT Image 2, given the photo in a chat in October 2026, returned a square with the room squeezed to fit. For a photo that will represent a real property on the MLS, use a purpose-built staging tool.

Why does ChatGPT change the room?

Two reasons. The older image models emit only fixed sizes (1024×1024, 1536×1024, 1024×1536), and in a chat even the newer one gave us a square, so a photo of any other shape gets re-framed. And it’s a general-purpose generative model built to produce a good image, not to reproduce the input with additions, so it treats architecture as something it can improve.

Which model does ChatGPT use for images now?

ChatGPT Images 2.5, launched on 8 September 2026 and available to all ChatGPT users, according to OpenAI. Developers get it as GPT-Image-2.5 Flare and Sunburst. It isn’t DALL-E, whatever some AI answers still say. We haven’t run our room test on Images 2.5 yet.

Is it against MLS rules to stage with ChatGPT?

Staging with any tool is fine if you disclose it. The problem is that MLS rules (Canopy, CRMLS and NorthstarMLS’s guidelines among them) separately prohibit altering walls, windows, doors, floors or dimensions. If the model moves a window, the image breaks the rule regardless of your disclosure.

Which AI is best for virtual staging?

In our test the Gemini image models preserved the room’s architecture and aspect ratio; the OpenAI models didn’t. Gemini 3.1 Flash Lite was both the cheapest at $0.0339 an image and the fastest at 6–8 seconds. Many purpose-built staging products are built on image-editing models of that type.

How much does AI virtual staging cost per photo?

The raw model cost is a few cents for the Gemini editors: we measured $0.0339 to $0.0678 an image. GPT Image 2 billed $0.237 for one run. What vendors charge is a different question, driven by monthly photo caps: VirtualStagingAI starts at $16 a month for six photos on annual billing ($25 month to month), and ApplyDesign’s auto mode works out at $10.50–$15 an image.

Can I just ask ChatGPT to keep the windows the same?

We did. The prompt explicitly listed the windows, frames, doorway, radiator, baseboards, floor and camera angle as things not to change. The windows changed anyway in all three OpenAI runs.

Free listing video maker

Staged the photos? Turn them into a listing video.

Reallyo turns your listing photos into a ready-to-post video with an AI script, voiceover and captions, plus social posts and flyers from the same listing. Free for one active listing.

Make a listing video free → See how it works

Sources

Checked 15 October 2026: OpenAI image models and size rules · OpenAI image generation guide · Introducing ChatGPT Images 2.5 · NorthstarMLS · Canopy MLS · CRMLS · VirtualStagingAI pricing · ApplyDesign pricing