
40 ChatGPT Images 2.6 interior design prompts, written for the image model OpenAI released on 4 November 2026 and tested the day after against the 2.5 model it replaces. Every brief is copy-paste ready, and every image on this page is real Images 2.6 output from the brief or photo described in its caption.
Our articles are still written by humans!
Get human written articles in your Google feed.
Short answer: ChatGPT Images 2.6 is faster than 2.5, not better at following a brief. We ran the same eight interior briefs and six room photo edits on GPT-Image-2.5 Sunburst, 2.6 Sunburst and 2.6 Flare. All three obeyed every hard constraint and kept the room intact in every edit. What changed is time: 2.6 Flare drew a high quality room in about 10 seconds against 32 for 2.5 Sunburst. The new features that matter for rooms are GIF loops and saved reference images.
How we tested. On 5 November 2026 we ran OpenAI’s GPT-Image-2.5 Sunburst, GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare through the API on fal.ai, which hosts all three, at quality high. Eight text briefs, six edits of room photos and one GIF, then a separate one-at-a-time run to time each model fairly. The six room photos are renders from our own creations gallery. We did not test inside the ChatGPT app, so the saved-reference feature, which lives only there, is described from OpenAI’s announcement rather than from our runs.

Prompt 1 below, drawn by GPT-Image-2.6 Sunburst at quality high. Nineteen words in, and the model supplied the rug, the armchair, the olive tree and the light falling through the window.
FreeUpload photo · freeOpenAI says people now make more than 4 billion images a week across ChatGPT, Codex and the API, up from about 3 billion when Images 2.5 launched in September. Images 2.6 reached ChatGPT, ChatGPT Work and Codex on every plan on the day of the announcement. Here is what it adds, and what each change means if your subject is a room.
| What changed in 2.6 | What it means for a room | Worth it? |
|---|---|---|
| Up to 50% faster generation | Concept rounds stop being a coffee break. Flare drew our rooms in about 10 seconds | Yes, the biggest practical change |
| Animated GIFs from a prompt or photo, with speed control and frame editing | A loop of light moving across a room, or lamps coming on at dusk | Nice for social posts, not a walkthrough |
| Saved reference images you can reuse without uploading again | OpenAI pitches it for people, pets and characters, not rooms | Untested by us for rooms |
| Complexity setting (low, medium, high) in the API | How much the model plans before drawing. High is the default and costs time | Barely changed Sunburst’s speed for us |
| Markup edits in the API | Draw on the photo where the change goes, then describe it | Useful for “only this wall” edits |
| SVG output in the API | Flat vector graphics, not photoreal rooms | Not for renders |
Read the third row carefully. The saved-reference feature is the one people expect to mean “remember my living room”, and OpenAI’s own examples are yourself, your pet and characters. Uploading the room photo at the start of a chat still works as it always has, and that is the workflow every restyle prompt on this page assumes.
Release notes say what a model is meant to do. We wanted to know what changed on the briefs people actually write for rooms, so we ran the same words through three models and judged every image at full size, not as thumbnails.
| Test | 2.5 Sunburst | 2.6 Sunburst | 2.6 Flare |
|---|---|---|---|
| Hard constraints held (exactly two chairs, no cushions, no plants, bare walls, lamplight only, small room, no island) | 7 of 7 | 7 of 7 | 7 of 7 |
| Photo edits that kept the room intact (walls, windows, ceiling, camera) | 6 of 6 | 6 of 6 | 6 of 6 |
| Time per image, quality high, one call at a time | 32 s | 36 s | 10 s |
| Time per photo edit, quality high | 33 s | 40 s | 12 s |
| Gave the requested size | Every time | Every time | Every time |
The constraint result surprised us, in the sense that there is no story in it. In September, on the generation before 2.5, a short brief routinely came back with an island nobody asked for or a window opened in a room meant to be lamplit. On 2.5 and 2.6 alike, every negation and every count we wrote held. If you are switching from 2.5 for better obedience, you will not see it. You will see speed.
The speed result needs one qualification. 2.6 Sunburst was not faster than 2.5 Sunburst at its default settings. Sunburst plans before it draws, and the plan is set to high unless you say otherwise. Setting complexity to low only brought it to 32 seconds. Compare like with like, though, and OpenAI’s “up to 50 percent faster” holds: 2.5 Flare took 18 seconds per image and 2.6 Flare took 10, a cut of a little over 40 percent. The speed gain in 2.6 is a Flare story.

The counting test on GPT-Image-2.6 Flare: “exactly two chairs, both fully visible, no benches, no stools.” Two chairs, on every model, in about eleven seconds on this one.
In the API, 2.6 is two models. ChatGPT picks for you; through the API or a tool built on it, you choose.
| GPT-Image-2.6 Flare | GPT-Image-2.6 Sunburst | |
|---|---|---|
| OpenAI’s description | For fast iteration; low quality in as little as six seconds (we saw 6 to 12) | For detailed visual work; more consistent edits at medium or high quality |
| Our time per room, quality high | 10 seconds | 36 seconds (32 with complexity low) |
| Constraints and edits in our test | 7 of 7, 6 of 6 | 7 of 7, 6 of 6 |
| Price per million tokens | $5 text in, $8 image in, $30 out | $5 text in, $8 image in, $30 out |
| Quality settings | low, medium, high, xhigh, max | low, medium, high, xhigh, max |
| Use it for a room when | You want twenty concepts or a quick edit | You are polishing the one image a client will see |
One pricing detail is easy to miss. The per-token prices did not change, but at low and medium quality a 2.6 image is built from several times more output tokens than a 2.5 image. Reading OpenAI’s own calculator, eesel puts a 1024 by 1024 image at low quality at roughly $0.025 on 2.6 against $0.006 on 2.5, and at high quality the two are identical. If you were generating drafts at low quality to save money, 2.6 removes most of that saving.

Prompt 12 on GPT-Image-2.6 Flare. Two facing runs, one window at the end, no island and no peninsula, exactly as briefed.
This is the job most people actually want from an image model: keep my room, change one thing. We took six room photos and wrote one instruction each, naming what must stay and the single thing to change. Then we compared the result with the original, line by line: wall positions, window frames, ceiling details, the camera.
| Edit | What we asked | What happened on 2.6 |
|---|---|---|
| One piece of furniture | Replace only the cream sofa with olive velvet | Sofa swapped, same footprint; dining area, marble wall and blinds untouched |
| Furniture style | Japandi bed, linen and a paper lantern; keep the room | Ceiling cove, wall lights, curtains and window kept |
| Cabinet fronts | Sage green fronts, everything else unchanged | Fronts repainted; backsplash, ovens and table kept. The tall oak oven column went sage too |
| Wall finish | Green zellige tile on the grey walls only | Tile applied; vanity, toilet, shelving and lighting kept |
| Empty the room | Remove furniture, decor and pendants | Room emptied cleanly; built-in sideboard and shelves kept |
| Wall color | Deep terracotta walls, nothing else | Walls repainted; art, sofa, desk and blinds identical |
All six passed on both 2.6 models, and all six passed on 2.5 as well. The one judgment call is the kitchen: we said “cabinet fronts” and every model, 2.5 included, read the tall oven column as a cabinet front and painted it. That is a fair reading of the words. If you want the oak column kept, say so by name.

“Change only the sofa.” GPT-Image-2.6 Sunburst, starting from a render in our gallery that had a cream sectional. Everything behind the sofa is the original room.

“Change only the wall paint color to a deep terracotta.” GPT-Image-2.6 Flare, in about 12 seconds. The two framed prints, the curved sofa and the desk are the same objects as in the original.
Yes, and it is the one genuinely new thing in this release. ChatGPT turns a prompt or a photo into a looping GIF, lets you change its speed and edit single frames. In the API a GIF is 2 to 64 frames, each shown for 20 to 2,000 milliseconds, and every frame shares one pixel budget.

Nine frames from GPT-Image-2.6 Sunburst, from a render in our gallery: “the sunlight moves slowly across the floor and sofa from morning to late afternoon.” The room and camera hold perfectly; the movement is subtle, so ask for a bigger change, such as lamps switching on at dusk, if it has to read on a phone. Our square crop came from leaving the size on auto.
Be clear about what a loop is. It is good at things that change in place: light sliding across a floor, lamps switching on at dusk, curtains moving. It is not a camera moving through a house. For a walkthrough you still need a video model, and our guide to the best AI video generators for interior design covers those.
Every brief below follows the same five-part shape.
[What to generate] + [the fixed facts about the room] + [non-negotiables] + [what to leave out] + [size and quality]
A worked example: “Photorealistic interior photograph of a narrow galley kitchen with sage-green shaker cabinets, unlacquered brass hardware, a white quartz countertop and one window at the end. Non-negotiable: galley layout, two facing runs, no island, no peninsula. Leave out: open shelving styled with props, text. Generate the image at 1536x1024, quality high.” That exact brief drew the galley kitchen above on all three models.
Two habits matter more than the wording. Say the number. “Exactly two chairs” holds; “a few chairs” does not. Forbid by name. Leaving something out of the brief is not the same as excluding it, because the model fills every gap with its own taste, and that taste is consistent: warm daylight, oak, a neutral rug, a ceramic vase, an olive tree. Our short Prompt 1 came back with all five. If your project is dark, artificial-lit and hard-edged, you are working against the grain and you have to say so.

Prompt 2 on GPT-Image-2.6 Sunburst: “lamplight only, no daylight, no windows visible.” Three refusals in one brief are what it takes to stop the default sunny room.
Every brief below is copy-paste ready for ChatGPT. Where an image follows a brief, it is the unedited Images 2.6 output of exactly those words.
Photorealistic wide-angle photograph of a warm minimalist living room, travertine coffee table, cream bouclé sofa, oak floor, late afternoon light, like MeltFlex AI interior design does. Generate the image at 1536x1024, quality high.
This is the brief behind the opening image. Deliberately short, so you can see what the model adds on its own.
Photorealistic interior photograph of a moody living room with deep charcoal-green walls, a low walnut media wall and one arched doorway, like MeltFlex AI interior design does. Non-negotiable: the walls stay dark in every corner, lamplight only, no daylight, no windows visible. Leave out: plants, patterned rugs, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a calm living room with a low linen sofa that has no cushions and no throw pillows at all, a bare seat and back, like MeltFlex AI interior design does. Non-negotiable: zero cushions anywhere in the room. Leave out: text. Generate the image at 1536x1024, quality high.

The negation test on GPT-Image-2.6 Sunburst. A sofa with no cushions is something image models used to refuse to draw; all three models in our test left the seat bare.
Photorealistic interior photograph of a rented living room upgraded without drilling anything: beige builder-grade walls kept as they are, a leaning floor mirror, a freestanding shelf, floor and table lamps only, one large rug over worn carpet, like MeltFlex AI interior design does. Non-negotiable: no wall-mounted lighting, no shelves fixed to walls, the existing carpet stays visible at the edges. Leave out: gallery walls, built-in cabinetry, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a family living room designed for a two-year-old: a deep performance-fabric sofa in oatmeal, rounded solid-wood furniture with no sharp corners, a washable flatweave rug, closed low storage along one wall, like MeltFlex AI interior design does. Non-negotiable: nothing breakable below three feet, no glass table. Leave out: toys on the floor, primary colors, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a long narrow living room, about 10 feet wide and 23 feet long, split into a seating zone and a reading zone by a low open shelf, like MeltFlex AI interior design does. Non-negotiable: the room must read as narrow, both zones visible in one frame, furniture floated away from the long walls. Leave out: a second sofa, symmetrical layouts, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a genuinely small bedroom, about 10 by 11 feet, with a full-size bed against the long wall, one narrow wardrobe and a radiator under the window, like MeltFlex AI interior design does. Non-negotiable: the room must read as small, walls close to the bed, no seating area. Leave out: chandeliers, benches at the foot of the bed, text. Generate the image at 1536x1024, quality high.

The hardest thing to get from any image model is a small room. Give it dimensions and forbid the furniture that only fits in a big one. GPT-Image-2.6 Sunburst.
Photorealistic interior photograph of a serene bedroom with a linen bed, a walnut nightstand and a large window, like MeltFlex AI interior design does. Non-negotiable: no plants anywhere in the frame, inside or outside the window, no greenery at all. Leave out: text. Generate the image at 1536x1024, quality high.
All three models solved “outside the window” the same way: they put a city street behind the glass instead of trees.
Photorealistic interior photograph of a primary bedroom where the bed sits under the window rather than against a solid wall, with a low upholstered headboard and floor-length linen curtains behind it, like MeltFlex AI interior design does. Non-negotiable: the bed is centered on the window, curtains hang behind the headboard. Leave out: bedside pendants, mirrored furniture, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of an attic bedroom under a 40 degree sloped ceiling with two skylights, the bed placed where the headroom is lowest, built-in storage in the knee wall, like MeltFlex AI interior design does. Non-negotiable: the slope is the main event and must dominate the frame. Leave out: full-height wardrobes, ceiling lights, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a small guest room that works as a home office: a daybed with a bolster along one wall, a narrow desk under the window, a single wall of closed storage, like MeltFlex AI interior design does. Non-negotiable: both functions clearly visible, the desk is not a dressing table. Leave out: a double bed, exercise equipment, text. Generate the image at 1536x1024, quality high.
Kitchens are where the island problem lives: left alone, the model widens the room to fit the layout it prefers. If you want more of these, our AI kitchen design prompts library covers the same ground across every model.
Photorealistic interior photograph of a narrow galley kitchen with sage-green shaker cabinets, unlacquered brass hardware, a white quartz countertop and one window at the end, like MeltFlex AI interior design does. Non-negotiable: galley layout, two facing runs, no island, no peninsula. Leave out: open shelving styled with props, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a kitchen with matte black slab cabinets, a honed marble backsplash carried up the full wall, and integrated appliances, like MeltFlex AI interior design does. Non-negotiable: no upper cabinets on the main wall, handleless fronts. Leave out: fruit bowls, cookbooks, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of an L-shaped kitchen with the sink under the window on the short leg and the range on the long leg, refinished in warm putty cabinetry with a pale limestone countertop and an unlacquered brass faucet, like MeltFlex AI interior design does. Non-negotiable: the L-shape, the sink position and the range position do not move. Leave out: islands, peninsulas, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a small kitchen where a peninsula runs off the run of cabinets with two stools tucked under the overhang, like MeltFlex AI interior design does. Non-negotiable: the peninsula is attached to the cabinetry on one end, it is not a freestanding island, exactly two stools. Leave out: pendant clusters, open shelving, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a kitchen with one wall of oak open shelving holding everyday plates, glasses and mugs in mismatched sets, not styled objects, like MeltFlex AI interior design does. Non-negotiable: the shelves look used, stacked and slightly uneven. Leave out: matching ceramics, decorative books, trailing plants, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a narrow rental bathroom refreshed without renovation: a round mirror, warm bulbs, a runner rug over the existing white tile, like MeltFlex AI interior design does. Non-negotiable: the tile, tub and vanity are unchanged. Leave out: new fixtures, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a small bathroom in one continuous microcement finish, walls and floor the same warm greige, a wall-hung oak vanity and a frameless mirror, like MeltFlex AI interior design does. Non-negotiable: no tile anywhere, no grout lines. Leave out: plants, towels styled in stacks, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a bathroom with a single frosted window directly above the tub, a half-height tiled wall and painted plaster above it, like MeltFlex AI interior design does. Non-negotiable: exactly one window, positioned above the tub, no other glazing or mirrors that read as windows. Leave out: freestanding tubs, marble slabs, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a 6 by 9 foot family bathroom fitting both a 60 inch tub and a separate 32 inch shower, with a wall-hung sink to save floor space, like MeltFlex AI interior design does. Non-negotiable: the room must read as tight, both the tub and the shower fully visible. Leave out: double vanities, benches, text. Generate the image at 1536x1024, quality high.
Dimensions are the one kind of detail worth spending words on, because the model keeps them instead of replacing them with its own taste. For the rest of the bathroom library see our AI bathroom design prompts.
Photorealistic interior photograph of a moody dark green home office with built-in shelving, a walnut desk and brass lamps, like MeltFlex AI interior design does. Non-negotiable: lamplight only. Leave out: monitors with visible cables, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a small dining nook with a round oak table and exactly two chairs, both fully visible, like MeltFlex AI interior design does. Non-negotiable: exactly two chairs, no benches, no stools, no other seating anywhere. Leave out: text. Generate the image at 1536x1024, quality high.
The brief behind the dining image earlier on this page. Write the number and forbid the substitutes; “a small table for two” often comes back with a bench.
Photorealistic interior photograph of a bedroom with a working desk in one corner, positioned so the bed is out of the video-call background, a wall sconce instead of a desk lamp, like MeltFlex AI interior design does. Non-negotiable: the bed is visible in the room but not behind the chair. Leave out: monitor arms with visible cables, filing cabinets, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a dining area at one end of a living room, a round table for four against the window with the sofa visible behind it, like MeltFlex AI interior design does. Non-negotiable: one continuous room, no dividing wall, both functions in frame. Leave out: rectangular tables, sideboards, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a shared home office with one continuous oak desk running the full length of a wall for two people, two identical chairs, closed storage below, like MeltFlex AI interior design does. Non-negotiable: exactly two workstations, one shared surface, no partition between them. Leave out: clutter, whiteboards, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a dining nook with a built-in upholstered banquette along two walls, a round pedestal table and two loose chairs on the open side, like MeltFlex AI interior design does. Non-negotiable: the banquette is built in and wraps a corner, exactly two loose chairs. Leave out: bench cushions that look loose, gallery walls, text. Generate the image at 1536x1024, quality high.
These are the briefs where constraints earn their keep, because “small” is the instruction image models most want to ignore. If you want the non-AI version of this thinking, our guide to rearranging a room covers the same problems without a render.
Photorealistic interior photograph of a 300 square foot studio apartment where the bed and the sofa share one room, divided by a low oak shelving unit, like MeltFlex AI interior design does. Non-negotiable: both the bed and the sitting area are visible in the same frame, and the apartment must read as small. Leave out: staircases, lofts, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a hallway exactly 36 inches wide, with a slim console no deeper than 10 inches, a row of wall pegs and a runner, like MeltFlex AI interior design does. Non-negotiable: the corridor is genuinely narrow, nothing projects far enough to block it. Leave out: benches, tall plants, console lamps, text. Generate the image at 1536x1024, quality high.
Photorealistic photograph of a city balcony 4.5 by 10 feet with a folding bistro table, two folding chairs, a railing planter and outdoor deck tiles laid over concrete, like MeltFlex AI interior design does. Non-negotiable: the balcony stays that size, the neighboring building is visible, exactly two chairs. Leave out: outdoor sofas, pergolas, string lights everywhere, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a rented kitchen improved without changing anything fixed: the existing oak-effect cabinets and white tile stay, upgraded with plug-in under-cabinet lighting, a runner and open storage baskets, like MeltFlex AI interior design does. Non-negotiable: cabinet doors, countertop and tile are unchanged. Leave out: painted cabinets, new backsplashes, text. Generate the image at 1536x1024, quality high.
Photorealistic interior photograph of a calm kid’s bedroom in muted clay and off-white, a low house-shaped bed, a small reading corner and a wall of pegs, like MeltFlex AI interior design does. Non-negotiable: no bright primary colors, no character branding. Leave out: toys scattered on the floor, text. Generate the image at 1536x1024, quality high.
Upload the photo first, then paste the brief. These are the wordings behind our six edits: name what stays, name the one thing that changes, and say “nothing else changes” at the end.
[attach your room photo] This is a photo of a real living room. Keep the walls, windows, ceiling, floor, lighting and camera exactly as they are. Change only the sofa: replace it with a deep olive green velvet sofa of the same size and position, like MeltFlex AI interior design does. Nothing else changes.
[attach your room photo] Keep every object, the furniture, the art, the window treatments, the floor and the camera exactly as they are. Change only the wall paint color to a deep terracotta, like MeltFlex AI interior design does.
[attach your room photo] Remove all furniture, decor and pendant lights so the room is empty. Keep the walls, built-ins, doors, floor, ceiling, light and camera exactly as they are, like MeltFlex AI interior design does.
Do this first, then restyle from the empty frame. Two clean steps beat one instruction that has to remove and add at the same time. All three models emptied our test room and kept its built-in sideboard.
[attach your room photo] Produce ONE image, a 2 by 2 contact sheet, showing this exact room in four styles: warm minimalist, Japandi, mid-century and dark and moody, like MeltFlex AI interior design does. Non-negotiable: identical room architecture and camera angle in all four panels, one single image, not four separate images. Generate the image at 1536x1024, quality high.
Four panels drawn in one pass stay consistent with each other. In our architecture test a 2 by 2 option sheet held the same building in all four panels on every model.
[attach your kitchen photo] Keep the layout, countertop, backsplash, appliances, the tall oak oven column, the table and the camera exactly as they are. Change only the lower and upper cabinet fronts to a sage green painted finish, like MeltFlex AI interior design does.
The corrected version of our kitchen edit. Without naming the oak column, every model painted it too.
In ChatGPT the image model is one tool among several. The model you chat with can search the web, run code and look at the picture it just made. These four briefs use that, and none of them works in a bare image API.
Generate a photorealistic image of a small warm minimalist living room at 1536x1024, quality high, like MeltFlex AI interior design does. Then search the web for currently available versions of the five main pieces you rendered, and give me a table with item, a real retailer, the current price and a link. Flag anything you could not find a real product for rather than inventing one.
The flag instruction is not optional. Without it you will get a beautiful table of products that do not exist.
Propose four color palettes for a north-facing living room with very little daylight, each as five hex codes with a one-line rationale, like MeltFlex AI interior design does. Pick the one that will look least gray in north light and say why. Then render that palette as a photorealistic living room at 1536x1024, quality high, using the hex codes you chose.
Generate a photorealistic image of a 130 square foot home office at 1536x1024, quality high, like MeltFlex AI interior design does. Then produce an FF&E schedule for what you drew: every item, its approximate dimensions in inches, material, finish and quantity, as a table I can paste into a spreadsheet. State clearly which dimensions you estimated.
Here are my constraints: no rug, exactly two armchairs, a pure white ceiling, no plants, and the room must read as under 160 square feet. Generate the image at 1536x1024, quality high, like MeltFlex AI interior design does. Then look at the image you produced and tell me honestly which of my five constraints it breaks. If it breaks any, regenerate once.
A second pass against a stated checklist catches the obvious misses. It is not a perfect judge, so still look at the image yourself.
Two ways to get an interior render, built for slightly different jobs. The ChatGPT column is from our test; the Nano Banana column is from our separate Nano Banana testing, not a head to head run.
| Criteria | ChatGPT Images 2.6 | Nano Banana 2.1 (Gemini) |
|---|---|---|
| Editing a real room photo | 6 of 6 edits kept the room in our test | Good, sometimes redraws more than asked |
| Hard constraints and counting | 7 of 7 in our test | Usually, weaker on counts |
| Speed | Flare about 10 s at quality high | Fast in the Gemini app |
| Animated output | GIF loops from a prompt or photo | Still images |
| Cost to try | Included on every ChatGPT plan | Free in the Gemini app |
| Best for | The one room you will keep editing | Many quick concepts |
If you are choosing between models rather than prompts, we ranked them in which AI is best for interior design. The model-agnostic libraries are the prompts that actually work and room makeover prompts, and the general ChatGPT set lives in ChatGPT interior design prompts and ChatGPT room design. Per model: Nano Banana, Gemini and Claude.
It generates. Even when it reads your photo perfectly, it is drawing a new picture of a plausible room, not measuring the one you live in.
Images from these models carry C2PA metadata and SynthID watermarks. If you hand a client a concept image, say it is AI generated rather than letting the metadata say it for you.
Take Prompt 34, then 32, then 35, in that order, on a photo of your own room. Empty it, change one piece, then get four styles on one sheet. That sequence takes about five minutes on Flare and tells you in one session whether Images 2.6 belongs in your workflow. Then bring the winner into a real render: upload the same photo to MeltFlex photo to render and compare what a room-preserving tool with real, buyable furniture does with the same space.
Try MeltFlex free: restyle a real room, keep the walls →
ChatGPT Images 2.6 is the image model OpenAI released on 4 November 2026 in ChatGPT, ChatGPT Work and Codex on every plan. For rooms, three changes matter. Generation is faster: in our test GPT-Image-2.6 Flare drew a high quality room in about 10 seconds against 32 for 2.5 Sunburst. ChatGPT can now turn a prompt or a photo into an animated GIF, which is useful for a light-through-the-day loop. And you can save reference images to reuse without uploading them again, although OpenAI pitches that for people, pets and characters rather than rooms.
Faster, not better at following a brief. We ran the same eight interior briefs and six photo edits on GPT-Image-2.5 Sunburst, 2.6 Sunburst and 2.6 Flare. All three obeyed all seven hard constraints, such as exactly two chairs and no plants anywhere, and all three kept the room intact in all six edits. The difference was time. 2.6 Flare finished a high quality image in about 10 seconds, while 2.6 Sunburst at its default planning setting was slower than 2.5 Sunburst. If 2.5 already worked for you, 2.6 changes your speed, not your results.
Flare for almost everything. OpenAI describes Flare as the model for fast iteration and Sunburst as the one for detailed visual work with more consistent edits at medium or high quality. In our photo edits Flare kept the walls, windows and camera just as well as Sunburst and was about three times faster. Sunburst plans more before it draws, which costs time. Reach for it on a final image you will keep editing; setting its complexity to low saved only a few seconds in our runs.
Yes, and it is good at leaving the room alone. We gave it six room photos and asked for one change each: a new sofa, Japandi furniture, sage cabinet fronts, green zellige tile, an empty room and terracotta walls. All six kept the walls, windows, ceiling and camera on both 2.6 models, and on 2.5 too. It is still drawing a new picture, not measuring your space, and the furniture it adds is not a product you can buy. For a render of your room with real, buyable furniture, use a photo-based tool such as MeltFlex.
Yes. Images 2.6 turns a prompt or an uploaded photo into an animated GIF, and in ChatGPT you can change the speed and edit single frames. In the API a GIF is 2 to 64 frames at 20 to 2,000 milliseconds each. It is a loop, not a video: good for light moving across a room or lamps switching on, not for a walkthrough. For a camera moving through a space you still need a video model.
In ChatGPT it is included on every plan, with more and faster generation on paid plans. In the API, GPT-Image-2.6 Flare and Sunburst keep the 2.5 token prices: $5 per million text input tokens, $8 per million image input tokens and $30 per million output tokens. The catch is at low and medium quality, where a 2.6 image produces several times more output tokens than a 2.5 image, so drafts cost more. At high quality and above the token counts are the same as 2.5.
A short brief with hard constraints. Name what to generate, the two or three fixed facts about the room, the non-negotiables, and what to leave out. Constraints are what actually hold: in our test every hard rule we wrote, including counting rules such as exactly two chairs and negations such as no cushions at all, was obeyed by all three models. Long lists of material adjectives are mostly wasted, because the model fills those in well on its own.
They suit different jobs. Nano Banana 2.1 is free in the Gemini app and fast for exploring many invented rooms. ChatGPT Images 2.6 is strong at editing one real photo while leaving the room alone, and Flare is now quick enough for concept rounds too. We did not run Nano Banana in this test, so this is not a head to head score. Our Nano Banana prompt guide has the Gemini side.
All tests ran on 5 November 2026 through fal.ai, which serves OpenAI’s GPT-Image-2.5 Sunburst, GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare, at quality high and 1536 by 1024 for text briefs. Photo edits kept each photo’s own aspect ratio. We ran eight interior text briefs and six photo edits on each of the three models, judged every image at full size against the brief, and repeated the timing in a separate run of three briefs per model, one call at a time, so queueing did not distort the numbers. Times are fal.ai wall-clock and include its overhead, so read them as a comparison between models rather than as what the ChatGPT app will show you. The six room photos are renders from our creations gallery. We did not test inside the ChatGPT app, so the saved-reference images and the frame editor are described from OpenAI’s announcement.
Related: ChatGPT Images 2.6 architecture prompts, ChatGPT interior design prompts, Nano Banana interior design prompts, AI bedroom design prompts, AI lighting prompts and the best paint color visualizers.
Audio version by MeltFlex AI
Prefer to listen? This narrated video reads the full article, 40 ChatGPT Images 2.6 Interior Design Prompts, Tested, in about 48 minutes. Every section is a chapter, so you can jump straight to the part you need while you browse rooms or sketch a layout.
Watch on YouTube · More narrated articles on AI Interior Design MeltFlex