GPT Image 2.6 Virtual Staging: Tested

TL;DR

Start with a clean empty-room photo. Specify the furniture to add and the architecture to keep, then compare the result with the source. Inspect every opening, boundary, shadow, and access path before using the image.

Direct answer: GPT Image 2.6 can visualize furniture in an empty room, but the output needs a photo check. State the camera and fixed architecture, request restrained furnishings, and reject any result that changes the property.

OpenAI released Images 2.6 on November 4, 2026, describing more editing control. It is available in ChatGPT, Codex, and the API; API developers can choose Flare or Sunburst. ChatGPT may present different controls, so do not assume an identical API-style model picker.

This guide is for an agent or real estate photographer starting with an empty listing photo. The comparison below documents one same-image, same-prompt test using a new synthetic empty room. The three additional staging prompt templates later in this article were not part of that test and remain untested examples.

One empty-room staging prompt across three image models

This is one documented case, not a benchmark ranking. The same new synthetic room and exact prompt went to GPT Image 2.6 Sunburst, GPT Image 2.5 Sunburst, and original Nano Banana. We kept each first result with no best-of selection.

GPT used high quality at 1536 × 1024. Nano exposed only 3:2 and returned 1248 × 832. There was no equivalent cross-model seed. Client wall time starts before request submission and includes input data-URI transfer, queueing, generation, five-second status polling, and result-JSON retrieval; it stops before the final image download and is not pure inference time.

Exact test prompt

Edit the attached empty living-room image into a restrained contemporary virtual staging concept.

CHANGE:
- Add exactly one cream two-seat sofa along the LEFT wall, one small round walnut coffee table, one light neutral rug, and one slim freestanding floor lamp.
- Use conservative furniture scale. Leave the blue door and the route to it unobstructed.

KEEP:
- Preserve both existing back-wall windows, every window division, the white radiator under the left window, the blue door on the right wall, all three ceiling downlights, the walls, skirting and honey oak floor.
- Keep the original camera position, field of view, crop, perspective and daylight. Do not add, remove or resize architectural features.
- Do not add wall art, built-ins, plants, extra furniture, text or logos.

Return one realistic edited image from the original view.

Parameters and recorded time

Model	Request controls	Returned image	Status	Client wall time
GPT Image 2.6 Sunburst	High quality, 1536 × 1024, one PNG	1536 × 1024	Success	59.42 s
GPT Image 2.5 Sunburst	High quality, 1536 × 1024, one PNG	1536 × 1024	Success	50.10 s
Nano Banana original	3:2, one PNG, limited generation	1248 × 832	Success	22.30 s

GPT Image 2.6 Sunburst

The result placed the cream sofa on the left wall and added the table, rug, and lamp. The blue-door route remained open. Both windows, radiator, three ceiling lights, floor, and main boundaries remained recognizable. This visual check does not establish pixel-level preservation or exact scale.

GPT Image 2.5 Sunburst

This result also used the left wall and included the table, rug, and lamp. The sofa and rug occupy more foreground; the radiator remains visible and the door route open. Books and a bowl appeared on the table although they were not requested. Main architecture stayed recognizable, not proven geometrically exact.

Nano Banana original

Nano added the four requested furnishings but placed the sofa against the back wall, not the left wall. It overlaps the window/radiator zone visually, though the door route remains open. The 1248 × 832 output also reframes the room, so pixel dimensions and controls are not equivalent.

What this case can and cannot show

Here, both GPT outputs followed the left-wall instruction and Nano did not. With n=1, no shared seed, and unequal controls, this cannot establish a generally better, faster, or more faithful model. Times are run records, not averages.

How to stage an empty-room photo with GPT Image 2.6

The safest workflow is source-photo first: define the property features, add a limited furnishing brief, then inspect the edited image against the original.

1. Choose a clean source photo

Use a sharp, level image that clearly shows the boundaries and openings. Avoid motion blur and severe lens distortion. Upload only a photo you can use.

2. List the fixed features and camera view

List what must not intentionally change: doors, windows, boundaries, ceiling line, built-ins, visible fixtures, and camera position. This reduces ambiguity but does not guarantee exact preservation.

3. Brief the furnishing plan

Name the room type, furniture group, style, and visual weight. Ask for conservative scale and clear access. Do not infer exact dimensions or fit from one photo.

4. Request an edited image

In ChatGPT, attach the photo and ask for an image edit. An API implementation can send the source image and prompt. The Flare model page documents an API model, not a universal ChatGPT control.

5. Compare the result with the original

Compare source and result side by side. Check the property before the decor. Revise or reject a changed opening, wall line, or blocked doorway.

Three GPT Image virtual staging prompt templates

These copy-ready prompts are authored examples. They have not been run as controlled GPT Image 2.6 tests, and they do not guarantee preservation or a usable result.

Empty living room

Edit the attached empty living-room photo into a restrained, contemporary virtual staging concept.

CHANGE:
- Add one appropriately scaled sofa, one compact coffee table, a low-pile rug, one floor lamp, and limited neutral decor.
- Use warm neutrals and light wood. Keep the room uncluttered.

KEEP:
- Keep the current camera position and field of view.
- Do not intentionally change the window count or position, doorway, walls, floor boundary, ceiling line, built-ins, radiator, or visible outlets.
- Do not add architectural features.

CHECK:
- Leave a clear route from the doorway into the seating area.
- Keep all window and door openings unobstructed.
- Use believable contact shadows and avoid furniture intersecting walls.

Dining room with visible circulation

Edit the attached vacant dining-room photo by adding a simple dining setup while keeping the property itself visually unchanged.

CHANGE:
- Add a modest dining table and chairs with a light, quiet style.
- Add one low centerpiece and a simple rug only if it does not crowd the room.

KEEP:
- Retain the camera view, wall and floor boundaries, windows, doors, trim, ceiling, and built-in elements.
- Do not invent an opening, remove a fixture, or change the room shape.

CHECK:
- Keep a visually clear walking route between every visible doorway.
- Do not place chairs through walls, trim, or each other.
- Keep furniture scale conservative; exact fit must be verified with real measurements.

Bedroom with conservative furniture scale

Virtually stage the attached empty bedroom as a calm, lightly furnished concept.

CHANGE:
- Add a bed with two small side tables, two lamps, and minimal bedding in soft neutral colors.
- Use restrained furniture proportions and leave visible floor around the bed.

KEEP:
- Preserve the intended camera view and do not intentionally alter windows, doors, closet openings, wall edges, ceiling line, flooring, or fixed fixtures.
- Do not add built-ins or structural details.

CHECK:
- Keep the doorway and closet access visibly clear.
- Check that bed edges and shadows do not merge into walls.
- Treat the image as a visualization, not evidence that a specific bed size fits.

What should you check before accepting a staged image?

Review architecture first, access and collisions second, then lighting and styling.

Decision	What you see	What to do
Accept	Doors, windows, boundaries, camera view, and fixed details visually match; furniture does not block access; shadows are plausible	Keep the candidate, then complete your normal listing and disclosure review
Revise	The property still matches, but one furnishing is too large, an access path is tight, or a shadow is distracting	Request one targeted change and repeat the keep list
Reject	A window or door changes, a wall moves, the camera shifts materially, a built-in disappears, or furniture collides with the room	Return to the source photo rather than editing the altered result

This visual review does not verify measurements, code, accessibility, or legal compliance. A convincing image can still imply the wrong furniture size or clearance.

How to correct a bad virtual staging result

Revise only the failed element and repeat the protected features.

Revise the last edited image using the original attached photo as the property reference.

Correct only this issue: [the sofa is too large / the doorway is blocked / the left window changed].

Keep every other accepted furnishing and style choice. Restore the original camera view, window and door positions, wall and floor boundaries, ceiling line, built-ins, and fixed fixtures. Do not add or remove architectural features. Return one corrected staged concept for another side-by-side check.

If a changed window or wall persists, reject the candidate and restart from the original.

When is a dedicated virtual staging workflow useful?

A dedicated workflow helps when the job is consistently “empty room in, furnished concept out.” Spaceora accepts JPG, PNG, or WebP up to 12 MB and presents room type, style, optional notes, generation, comparison, and download controls. You still need to inspect features, scale, and disclosure obligations.

Spaceora and GPT Image 2.6 are not the same product.

A practical final check

Use virtual staging as visualization, not dimensional evidence. Start from the original, keep the furnishing plan limited, and reject any output that changes the home.

FAQ