GPT Image 2.6 AI Interior Design: Tested

TL;DR

Use GPT Image 2.6 AI interior design to explore one direction at a time from an occupied-room photo. List what stays and what changes, then compare concepts. Verify dimensions, products, materials, availability, and installation in the real room.

Direct answer: Upload a clear room photo, list what stays, and request one coherent change. Treat the result as a concept, not a shopping plan or proof of fit. Revise one variable per round.

OpenAI introduced Images 2.6 on November 4, 2026, emphasizing editing control. It is available through ChatGPT, Codex, and the API; API developers can choose Flare or Sunburst. ChatGPT may expose different controls.

This workflow is for a furnished room where the homeowner wants to retain selected pieces. It is not virtual staging for a vacant property. The comparison below documents one same-image, same-prompt test using a new synthetic furnished room. The three additional interior-design prompt templates later in this article were not part of that test and remain untested examples.

One occupied-room textile edit across three image models

One new synthetic furnished-room input and the exact prompt went to three models. We retained each first successful image; Nano first returned HTTP 422/no image, then one manual same-input, same-prompt retry produced the image below.

GPT used high quality at 1536 × 1024; Nano used 3:2 and returned 1248 × 832; no equivalent seed. Client wall time began before submission and covered input data-URI transfer, queueing, generation, five-second status polling, and result-JSON retrieval; it stopped before final image download and is not pure inference.

Exact test prompt

Edit the attached furnished living-room image as a targeted textile refresh.

CHANGE ONLY:
- Replace the rust patterned rug with a light cream low-pile rug in the same footprint.
- Replace the burgundy cushions with muted sage green cushions, retaining the same cushion count and placement.
- Replace the dark curtains with light beige linen curtains using the same curtain rail and approximate hanging positions.

KEEP:
- Keep the mustard yellow sofa exactly as the same sofa: shape, upholstery color, arms, three seat cushions and position.
- Keep the rectangular walnut coffee table: shape, four legs, wood color, size and position.
- Preserve the three-panel white-framed window and its divisions, radiator, open right-hand doorway, air conditioner, walls and walnut floor.
- Keep camera position, field of view, crop, perspective and daylight unchanged.
- Add no people, new objects, text or logos. Do not renovate or enlarge the room.

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	60.37 s
GPT Image 2.5 Sunburst	High quality, 1536 × 1024, one PNG	1536 × 1024	Success	51.76 s
Nano Banana original	3:2, one PNG, limited generation	No image, then 1248 × 832	First request HTTP 422; same-input, same-prompt manual retry succeeded	21.61 s failed attempt + 21.81 s successful retry

The 21.61-second failure and 21.81-second retry are separate attempt times, exclude the human interval, and must not be summed as 43.42-second end-to-end or model time.

GPT Image 2.6 Sunburst

The rug became cream, three cushions became green, and the curtains beige. The mustard sofa, walnut table, three-panel window, radiator, doorway, and air conditioner remained recognizable. Rug footprint, cushion positions, and table proportions are visually close, not demonstrably exact.

GPT Image 2.5 Sunburst

GPT Image 2.5 also completed all three textile changes and retained three green cushions. The sofa, four-legged table, window divisions, radiator, doorway, air conditioner, walls, and floor remain visible. Curtain folds, cushion shapes, rug edges, and small furniture details were regenerated, not proven identical.

Nano Banana original

The retry changed all three textiles and retained the main room elements. Its 1248 × 832 framing shifts more visibly: the sofa fills more of the left side, the table appears shorter, and the window/door composition moves relative to the source. That matters when crop and perspective should stay unchanged.

What this case can and cannot show

All successful outputs made the three changes. Nano's first request returned HTTP 422 after 21.61 seconds before an identical manual retry succeeded. With n=1, unequal controls, and no shared seed, this cannot rank quality, reliability, or speed. No billing response means no actual-price claim.

Redesign your room photo in five steps

The key is to make the design brief smaller before asking the model to make the room different.

1. Choose one useful view

Use a level photo showing the main furniture, openings, and two walls. Avoid crops that hide circulation and filters that change color. Upload only a photo you can use.

2. List what stays

Name the fixed architecture and furniture to keep. Be concrete: “keep the blue sofa, oak table, left window, fireplace, and camera view.” This guides rather than guarantees preservation.

3. Specify one coherent change

Choose a connected decision, such as a warmer palette with new textiles. Avoid changing the palette, layout, furniture, and architecture in one round.

4. Generate a concept

Attach the photo in ChatGPT and request an edit, or use an API implementation that accepts the image and prompt. The Flare page documents an API model, not a universal ChatGPT interface.

5. Revise one variable

Keep accepted parts and change one element: wall color, rug, pendant, or chair. Check the full image because other areas can change unexpectedly.

Build a keep/change brief before prompting

A useful brief separates constraints from preferences. “Budget-friendly” guides a visual direction; it does not verify price or availability.

Brief field	Filled example for a furnished living room
Fixed architecture	Keep the window, doorway, fireplace, wall boundaries, ceiling line, and camera view
Existing furniture	Keep the blue sofa and oak coffee table; do not replace or resize them intentionally
Renter limits	No built-ins, wiring changes, floor replacement, or permanent wall treatment
Change	Use a warmer neutral wall color, lighter curtains, a smaller rug, and one floor lamp
Lighting	Keep the daylight direction; add plausible warm lamp light without turning day into night
Color and material	Warm off-white, muted rust accent, natural linen, light wood; avoid glossy marble
Budget instruction	Prefer visually simple, widely available categories; verify real products and costs later

Three GPT Image 2.6 interior design prompt templates

These prompts are original, untested templates. They describe what to request and inspect; they are not evidence of model performance.

Keep the sofa and change the palette

Edit the attached occupied living-room photo as a design concept.

KEEP:
- Keep the current blue sofa and oak coffee table as the same intended pieces.
- Keep the window, doorway, wall and floor boundaries, fireplace, ceiling line, and camera view.

CHANGE:
- Shift the palette to warm off-white, muted rust, and natural linen.
- Replace only the rug, curtains, cushions, and floor lamp with restrained alternatives.
- Do not add built-ins or change the room layout.

CHECK:
- Confirm the sofa and table remain visually consistent with the source.
- Keep circulation clear and shadows aligned with the source daylight.
- Return a concept image, not a product list or measurement claim.

Make a small room feel calmer without blocking circulation

Edit this furnished small-room photo to create a lighter, calmer concept while keeping the existing architecture and main seating.

KEEP:
- Keep the sofa, window, door swing area, radiator, wall boundaries, flooring, and camera view.

CHANGE:
- Use a pale neutral palette, one compact side table, a visually lighter rug, and limited wall decor.
- Remove visual clutter only where it is not a fixed feature or required furniture.

CHECK:
- Keep a clear visual route between the doorway and window.
- Do not imply that a new item fits; real dimensions must be checked.
- Reject any concept that moves an opening or materially changes the room shape.

Create a renter-friendly dining corner

Edit the attached occupied dining-corner photo as a renter-friendly design direction.

KEEP:
- Keep the dining table, existing flooring, walls, window, doorway, ceiling, and camera position.
- No construction, built-ins, rewiring, new openings, or permanent surface changes.

CHANGE:
- Add removable curtains, a plug-in floor lamp, seat cushions, a small washable rug, and non-permanent wall art.
- Use warm, quiet colors that work with the existing table.

CHECK:
- Keep chairs and access paths visually clear.
- Keep the daylight direction believable.
- Treat colors and materials as visual references that require real samples.

Compare concepts without losing the room

Compare concepts that answer the same decision. Keep furniture fixed while comparing two palettes. If several variables change together, you cannot tell which choice helped.

Decision area	What to compare	What still needs verification
Light	Does the palette work with the visible daylight direction?	Paint and textile samples in the real room
Function	Does the concept support how the room is used?	Daily habits, storage, maintenance, and installation
Circulation	Are doors and walking routes visually clear?	Real clearances and furniture measurements
Existing pieces	Do retained items still look like the intended pieces?	Exact dimensions, condition, color, and model identity
Renter feasibility	Are changes removable and non-structural?	Lease terms and landlord approval

Use the matrix to choose a direction, not to skip measurements or approval.

Move from an AI concept to real decisions

Measure the real room, not the generated image. Record relevant wall lengths, clearances, ceiling height, and retained furniture, then compare them with product specifications.

Check local product availability and real samples. Screens and generated images render colors differently, so inspect paint, fabric, flooring, and wood under the room's actual light.

Spaceora is separate and is not presented as a GPT Image 2.6 integration.

Spaceora accepts JPG, PNG, or WebP up to 12 MB and presents room type, style, optional notes, generation, comparison, and download controls.

A practical next step

Write the keep list before choosing a style. Make one coherent change, compare it with the original, then move to measurements, samples, availability, and installation questions.

FAQ