
Midjourney is now the most widely adopted AI image tool in architecture practices, and for one reason: it turns a sentence into competition-grade atmosphere in about a minute. But most architects write prompts like "modern house, nice render" and get generic stock imagery back. The output quality is almost entirely a function of how you structure the prompt.
This guide is built for architects, not decorators. Every prompt below targets buildings, facades, massing, sections, site context, and client presentation, with the exact materials, architect references, and parameters that produce something you can actually put in front of a client. All 45 are copy and paste ready for Midjourney V8.1, the current default model, and they still run on V7 with one small change noted below. If you are looking for room and furniture prompts instead, those live in our ChatGPT interior design prompts and Gemini interior design prompts guides, and for prompts that are not tied to any one tool see our AI render prompt library.

Every prompt below was run in Midjourney V8.1 before it went on this page, not adapted from a V6 list and relabelled. That matters more than usual this year, because the raw flag changed syntax between the V7 and V8 lines and a large share of the architecture prompt lists still circulating silently break on it. Where a prompt needed different parameters to hold up under V8.1, the parameters here are the ones that worked, not the ones that read well.
I run design and rendering at MeltFlex, where the day job is the opposite problem to Midjourney: keeping a real building intact rather than inventing a plausible one. That is why this article is blunt about what Midjourney will not do for a project you have already modelled, and why the comparison further down sends you elsewhere for model-accurate work instead of overselling a text-to-image tool. Version claims here are dated to the 10 June 2026 V8.1 default; if you are reading this much later, check the parameters against the current release before trusting them.
Midjourney moved fast in the first half of 2026. V8 landed in alpha on 17 March, and V8.1 became the default model on 10 June 2026. If you saved a set of architecture prompts before that, they still work, but two of them are now written the old way and one new flag is worth adding to every board render. Here is the short version.
| What | V7 | V8.1 | Why an architect cares |
|---|---|---|---|
| Raw mode | --style raw | --raw | Same job, new spelling. This is the flag that stops Midjourney softening edges and inventing decoration on a facade. |
| Model flag | --v 7 | --v 8 | You can drop it entirely now that V8.1 is default, but leaving it in makes a prompt library reproducible. |
| Resolution | Upscale step | --hd | Native 2K straight out of the render, so a competition board image no longer needs an upscale round trip. |
| Coherence | n/a | --q 4 | Extra coherence pass. Worth it on structural frames and repetitive facades where V7 used to drift. |
| Speed | Baseline | ~4 to 5x faster | A 30 option massing study now fits inside a coffee break rather than an afternoon. |
Your V7 personalization profiles, moodboards and style references all carry over, so nothing you built up last year is wasted. One cost note worth knowing before you plan a big board run: style reference and moodboard jobs run about 4x slower and cost about 4x a regular job, so use them deliberately rather than on every test render.
Every prompt in this guide is written in the V8 form. To run any of them on V7 instead, swap --raw --v 8 back to --style raw --v 7 and change nothing else.
Where these version facts come from
The release dates, the new parameters and the backwards-compatibility and cost notes above are taken from Midjourney’s own V8 alpha announcement and its style reference documentation, not from secondhand roundups. Checked 25 July 2026.
One thing we are deliberately not certain about: --oref shipped as a V7 feature and support across the V8 line has been rolling out rather than landing all at once. We have flagged that in the omni reference section instead of stating it works everywhere, because a prompt library that quietly breaks on your default model is worse than one that tells you to test first. If you find it behaving differently, the parameter, not this guide, is the thing that moved.
Before the prompts, learn the structure. Every reliable architecture prompt follows the same six-part order. Keep this order and Midjourney reads it the way an architect would describe a project.
[View type] + [building type and style] + [key materials] + [site and context] + [time of day and lighting] + [mood and render style] + [parameters]
The two levers that separate amateur output from professional output are materials and light. "Concrete house" gives you a generic box. "Board-formed concrete with bronze-framed glazing and a zinc standing-seam roof" gives you something you can discuss with a client. Naming a reference architect (Tadao Ando, Kengo Kuma, SANAA, BIG, Snohetta) pulls the whole image toward a coherent language.
| Parameter | What it does | Use for architecture |
|---|---|---|
| --ar 16:9 | Wide aspect ratio | Exterior perspectives, landscape shots |
| --ar 3:2 / 4:3 | Hero / board ratio | Client hero renders, presentation boards |
| --ar 2:3 | Tall portrait | Towers, sections, full-height facades |
| --raw | Less auto-stylisation | Cleaner geometry and structural lines (was --style raw on V7) |
| --hd | Native 2K output | Board and print imagery with no upscale round trip |
| --q 4 | Extra coherence pass | Repetitive facades, structural frames, anything that used to drift |
| --sref | Style reference image | One visual language across a whole competition board |
| --stylize 50 to 250 | Artistic license | Low for accuracy, high for mood imagery |
| --chaos 0 to 50 | Variety in the first grid | Early concept and massing exploration |
| --no people, cars, text | Negative prompt | Strip clutter from clean facade studies |
| --v 8 | Model version | Best current quality for buildings |
The formula is easier to trust when you see what it costs you to ignore it. Both images below describe the same brief, a concrete house on a wooded slope. The only difference is how the prompt was written.
The weak prompt: "modern concrete house in a forest, nice render"

You get a plausible building and nothing you can defend in a meeting. The concrete has no formwork, the glazing has no frame system, the light has no direction and no time of day, and the massing is whatever the model felt like. It is stock imagery.
The strong prompt: "Exterior of a single-family house on a sloped forest site, board-formed concrete with visible timber grain and deep reveals, bronze-framed glazing, cantilevered upper volume, low golden hour side light raking across the concrete, calm and monolithic, photorealistic architectural visualization --ar 16:9 --raw --no people, cars --v 8"

Same brief, same model, roughly the same number of words. The difference is that every word is doing a job: a named formwork technique, a named frame metal, a structural move, a light direction and a mood. That is a render you can put on a board.
If you only change one habit after reading this guide, make it this one. Swap adjectives ("modern", "beautiful", "stunning") for nouns a contractor would recognise. Materials and light are the two levers that carry almost all of the quality.
About these two images
Both comparison renders above were generated with our own engine, not Midjourney, and we would rather say so than let you assume otherwise. We did it that way deliberately: running both prompts through one model, back to back, isolates prompt structure as the only variable. If we had shown two Midjourney jobs, you could fairly ask how much of the difference was the model rerolling rather than the wording.
The point being demonstrated is tool-neutral. Naming a real formwork technique, a real frame metal and a light direction improves the output of any image model, which is why the same 5-part structure shows up in our tool-neutral render prompt library. The 45 prompts themselves are written for Midjourney and its parameters, and the V8 syntax in them is what you should paste into Midjourney.
Disclosure: MeltFlex is our product. We mention it below alongside Midjourney, Veras and Enscape because we think the honest position is that it does a different job (interiors from a real photo) rather than a better one.
"Wide-angle exterior of a two-storey modernist house, flat roof, floor-to-ceiling glazing, white render and warm oak cladding, cantilevered upper volume, on a flat lawn with mature trees, soft overcast daylight, calm and refined, photorealistic architectural visualization --ar 16:9 --raw --v 8"
"Exterior of a single-family brutalist villa, board-formed concrete with deep reveals, bronze-framed windows, monolithic massing, on a sloped forest site, golden hour side light, heavy and sculptural, photorealistic architectural render, in the style of Tadao Ando --ar 16:9 --raw --no people, cars --v 8"
"Eye-level exterior of a low Japanese minimalist house, charred shou-sugi-ban timber cladding, deep eaves, paper-screen glazing, gravel garden and a single maple tree, quiet morning light, serene and restrained, in the style of Kengo Kuma --ar 3:2 --raw --v 8"
"Exterior of a compact Nordic cabin, blackened timber cladding, gabled roof, large picture window, on a rocky coastline with low pine, cold blue overcast light, isolated and atmospheric, photorealistic --ar 16:9 --raw --v 8"
"Exterior of a single-storey glass pavilion house, steel frame, full glass curtain wall, travertine plinth, flat roof with thin slab edge, on a manicured lawn at dusk with warm interior glow, elegant and transparent, in the style of Mies van der Rohe --ar 16:9 --v 8"
"Exterior of a contemporary passive house, larch timber cladding, triple-glazed windows, green sedum roof, solar panels integrated flush, on a meadow site, bright clear daylight, healthy and grounded, photorealistic architectural visualization --ar 16:9 --raw --v 8"
"Exterior of a modern Mediterranean villa, white lime plaster, arched openings, terracotta accents, infinity pool, olive trees and cypress, warm late-afternoon sun, luxurious and serene, photorealistic --ar 3:2 --raw --v 8"
"Dramatic exterior of a house cantilevered over a steep hillside, exposed concrete and dark steel, glazed living volume projecting into the valley, low cloud below, moody overcast light, bold and engineered, photorealistic architectural render --ar 16:9 --raw --v 8"
"Exterior of a contemporary art museum, white perforated metal facade, fluid curved volumes, reflecting pool at the entrance, urban plaza with people for scale, soft diffused daylight, light and weightless, in the style of SANAA --ar 16:9 --raw --v 8"
"Street-level exterior of a corporate headquarters tower, double-skin glass curtain wall, vertical aluminium fins, landscaped forecourt, dusk with glowing interiors and city lights, sleek and corporate, photorealistic architectural visualization --ar 2:3 --v 8"
"Exterior of a boutique hotel, warm brick and bronze detailing, rhythmic arched windows, ground-floor cafe spilling onto the pavement, evening with warm lighting, inviting and crafted, photorealistic --ar 3:2 --raw --v 8"
"Exterior of a public library and cultural centre, timber and glass, sloping accessible roofscape people can walk on, civic plaza, bright Nordic daylight, open and welcoming, in the style of Snohetta --ar 16:9 --raw --v 8"
"Exterior of a mixed-use development, masonry podium with active retail frontage and a slender residential tower above, planted terraces, busy city context, clear morning light, dense and urban, photorealistic architectural render --ar 2:3 --raw --v 8"
"Exterior of a futuristic transport hub, sweeping white parametric roof shell, exposed steel ribs, large glazed concourse, travellers for scale, bright daylight, dynamic and structural, in the style of Zaha Hadid --ar 16:9 --v 8"
"Exterior of a converted brick warehouse, retained heritage facade with a new black steel and glass rooftop extension, urban canal-side context, overcast afternoon, raw and contemporary, photorealistic architectural visualization --ar 3:2 --raw --v 8"

This is where Midjourney earns its place. Use these at the very start, before you commit to a direction, and turn up --chaos to widen the spread of options.
"Architectural massing model, plain white volumes, no materials, clean geometric blocks on a white base, soft studio lighting, neutral grey background, conceptual and abstract, --ar 4:3 --style raw --chaos 25 --v 8"
"Architectural concept that blends a loose hand sketch with a photorealistic render, half pencil linework half finished building, contemporary museum, on white paper, creative and exploratory, --ar 3:2 --v 8"
"Close-up facade study, parametric perforated metal screen, repeating algorithmic pattern, raking afternoon light casting deep shadows, abstract and detailed, in the style of Zaha Hadid --ar 1:1 --raw --v 8"
"Exploded axonometric diagram of a small building, separated layers for structure, facade, and roof floating apart, clean white and grey, thin black outlines, on white background, technical and clear, --ar 4:3 --raw --v 8"
"Architectural concept collage, mixed media, photographed materials, scaled human figures, trees and sky cut-ins, hand-drawn annotations, contemporary and editorial, --ar 3:2 --v 8"
"Hero exterior of a competition-winning cultural building, bold sculptural form, dramatic cantilever, civic plaza at blue hour with crowds and warm interior glow, cinematic and aspirational, photorealistic architectural visualization --ar 16:9 --raw --v 8"
These imitate the language of technical drawings. They are perfect for boards and storytelling, but remember they are not measured. For anything buildable, draw it in CAD.
"Cutaway section perspective of a three-storey house, the building sliced open to reveal furnished interiors and structure across floors, warm interior lighting, people inside for scale, illustrative and detailed, --ar 16:9 --raw --v 8"
"Isometric cutaway of a small apartment block, one corner removed to show stacked units, clean and colourful, soft shadows, on white background, architectural illustration, --ar 1:1 --raw --v 8"
"Worm's eye axonometric of a concrete structural frame, looking up at columns, beams, and slabs, monochrome grey, dramatic perspective, on white background, structural and bold, --ar 2:3 --raw --v 8"
"Architectural wall section render, layered build-up of cladding, insulation, structure, and interior finish, labelled illustrative style, warm material colours, on a light background, technical and clean, --ar 2:3 --raw --v 8"
"Coloured program diagram of a mixed-use building in axonometric, each function a different soft colour block, retail, office, residential, clean and editorial, on white background, --ar 4:3 --raw --v 8"
"Cross-section of a hillside house showing the building cut against the slope, retaining structure, terraced levels, sky above and earth below, people and trees for scale, illustrative architectural section, --ar 16:9 --raw --v 8"
"Aerial bird's eye view of a new neighbourhood masterplan, low-rise blocks around green courtyards, tree-lined streets, a central plaza and water feature, bright daylight, organised and green, photorealistic --ar 16:9 --raw --v 8"
"Eye-level streetscape of a revitalised urban high street, mixed-height buildings, active ground floors, wide pavements, cyclists and pedestrians, warm afternoon light, lively and human, photorealistic architectural visualization --ar 16:9 --v 8"
"Landscape architecture render of a public plaza, patterned stone paving, integrated planting and seating, water jets, surrounded by contemporary buildings, golden hour, social and vibrant, --ar 3:2 --raw --v 8"
"Eye-level view of a contemporary infill building inserted between two historic stone buildings, respectful modern facade in zinc and glass, narrow European street, soft overcast light, contextual and confident, photorealistic --ar 2:3 --raw --v 8"
"Aerial view of a waterfront cultural district, sculptural buildings along a promenade, boardwalks and planting, boats on the water, blue hour with reflections, cinematic and expansive, photorealistic architectural render --ar 16:9 --raw --v 8"
Run these to lock material and light direction before you commit. Keep the building simple so the study reads clearly.
"Close-up material study of a board-formed concrete wall, visible timber grain and tie holes, raking morning light revealing texture, minimal context, tactile and honest, --ar 1:1 --raw --v 8"
"Facade study in weathered corten steel, rich rust tones, recessed glazing, golden hour side light, warm and industrial, photorealistic architectural detail --ar 3:2 --raw --v 8"
"Facade study of a glass curtain wall at dusk, layered reflections of sky and city, warm interior glow behind the glass, crisp and luminous, photorealistic --ar 2:3 --v 8"
"Facade study combining handmade brick and warm timber screening, soft late-day light, human-scale entrance, crafted and welcoming, photorealistic architectural detail --ar 3:2 --raw --v 8"
"Facade study in travertine stone and bronze mullions, refined proportions, calm shadow play, soft daylight, luxurious and timeless, photorealistic --ar 1:1 --raw --v 8"
"Night exterior of a contemporary building, dark facade dissolving into shadow, glowing interior light spilling from large openings, wet pavement reflections, dramatic and cinematic, photorealistic architectural visualization --ar 16:9 --v 8"

These are architectural interiors, the atriums, stairs, and structural spaces that define a building, not soft furnishing. If you want furniture, colour, and decor prompts, use our Nano Banana interior design prompts instead.
"Interior of a double-height atrium, exposed concrete and timber, skylight flooding the space with daylight, floating walkways, people for scale, monumental and calm, photorealistic architectural visualization --ar 2:3 --raw --v 8"
"Interior view of a sculptural cantilevered concrete staircase, spiralling upward, soft top light, minimal white surroundings, elegant and structural, photorealistic --ar 2:3 --raw --v 8"
"Interior of a cathedral-like cross-laminated timber hall, exposed glulam structure, rhythmic beams, clerestory daylight, warm and uplifting, photorealistic architectural render --ar 16:9 --raw --v 8"
"Interior of a white museum gallery, polished concrete floor, hidden cove lighting, tall ceilings, a few visitors, quiet and refined, photorealistic architectural visualization --ar 3:2 --raw --v 8"
"Interior of a brutalist office lobby, raw concrete walls and coffered ceiling, dramatic shafts of daylight, reception desk in bronze, imposing and refined, photorealistic --ar 16:9 --raw --v 8"
"Photorealistic hero render of a finished residential building at dusk, warm interior lighting, a family arriving home, landscaped garden, cinematic depth of field, aspirational and warm, architectural marketing visualization --ar 3:2 --v 8"
"Drone-style aerial view of a completed development, buildings and landscape together, long late-day shadows, surrounding context, polished and aspirational, photorealistic architectural marketing render --ar 16:9 --raw --v 8"
Add any of these to the end of a base prompt to shift the result without rewriting it.
This is the part that separates a competition board from a folder of nice pictures. Fifteen renders that each look great but share no palette, no light and no drawing language read as fifteen different projects. Midjourney has two tools for fixing that, and most architects use neither.
A style reference carries colour, mood, lighting and rendering style from one image to the next without copying the subject. Generate or pick your single best render, take its URL, and append it to every other prompt on the board:
"Aerial masterplan of the same scheme, low-rise blocks around green courtyards, bright daylight, photorealistic --ar 16:9 --raw --sref [your-hero-image-url] --v 8"
The masterplan will now come back in the same palette and the same light as your hero shot, even though the viewpoint and the content are completely different. You can pass several URLs separated by spaces to blend a few sources. Remember the cost note from earlier: --sref jobs run around 4x slower and cost around 4x a normal job, so lock your reference image first and only then run the board.
Omni reference goes further and holds an actual subject across generations, not just a style. For architects that means keeping the same building recognisably itself while you move the camera, the season or the time of day. The syntax pairs the reference with a weight: --oref [image-url] --ow 200. The weight runs from 1 to 1000, and the useful architectural range sits between roughly 100 and 400, low enough that the model can still light the scene properly, high enough that the massing does not drift.
Two honest caveats. Omni reference arrived as a V7 feature and support across the V8 line has been rolling out rather than landing all at once, so test it against your current default before you plan a whole board around it. It is also incompatible with some faster modes, including --q 4, so you cannot stack it with the extra coherence pass.
--chaos 25 and no references until one image has the palette and light you want.--sref pointing at it, adding --oref if you need the building itself to stay consistent.--hd for native 2K.That sequence is why V8 matters more than the version number suggests. When renders come back 4 to 5 times faster, the exploration phase stops being the expensive part, and you can afford to throw away far more options before you commit.
Midjourney is the best tool for concept imagery, but it is not the only tool, and it is not the right one for every job. Here is how it sits next to the alternatives architects reach for.
| Tool | Price | Best for | Weakness |
|---|---|---|---|
| Midjourney | From $10/mo | Concept, mood, competition imagery | Hallucinates geometry, no model input |
| Veras | Subscription | AI render inside Revit, Rhino, SketchUp | Less artistic range than Midjourney |
| Enscape / Twinmotion | Subscription | Accurate real-time renders from your model | Not generative, slower for ideation |
| MeltFlex | Free tier | Photoreal interior renders from a real photo | Interiors and rooms, not exteriors |
The honest workflow most practices land on: ideate exteriors and atmosphere in Midjourney, develop the real geometry in BIM, render accurately in Enscape or Veras, and visualise the interior fit-out from a real photo in a model-aware tool. Midjourney is the spark at the front, not the deliverable at the end.
A strong architecture prompt follows a fixed order: view type, building type, architectural style, key materials, site and context, time of day and lighting, mood, render style, then parameters. Example: wide-angle exterior of a three-storey concrete house, brutalist, board-formed concrete and bronze-framed glazing, on a sloped forest site, golden hour, calm and monolithic, photorealistic architectural visualization --ar 16:9 --raw --v 8. The more concrete the materials and lighting, the more usable the result. Vague words like modern or beautiful waste tokens.
V8.1, which became the default on 10 June 2026, is now the one to use. It follows prompts more closely, holds small details better, renders roughly 4 to 5 times faster than V7, and adds --hd for native 2K output with no upscale step. The one thing to relearn is the raw flag: V7 wrote it as --style raw, and the V8 line takes it as --raw. Raw still does the same job for architects, cutting the automatic stylisation that softens edges and invents decorative elements, which keeps facades and structural lines clean. For competition and mood imagery, drop raw and let the default aesthetic run.
No, and you should not present it as one. Midjourney generates concept imagery from text, so it hallucinates geometry, invents structural members, and cannot read your actual model dimensions. It is excellent for early concepts, mood boards, competition entries, and client-facing atmosphere, but it does not produce accurate renders of a specific building you have modelled. For accuracy you still need Enscape, V-Ray, Twinmotion, or a model-aware AI render tool such as photo to render. Use Midjourney upstream of those, not instead of them.
Most use it at the front of the process. Common uses are exploring massing and facade options before committing, generating mood and material direction for a pitch, producing competition hero imagery, and creating atmosphere visuals for early client meetings. A typical session generates 15 to 30 variations in under an hour, then the architect narrows to two or three directions and develops those in BIM and dedicated render software. Around 44 percent of architects now use AI for concept images, so this is standard practice, not a gimmick. We broke the time saving down in how much time AI actually saves architects.
Four do the heavy lifting. Aspect ratio: --ar 16:9 for exterior perspectives, --ar 3:2 for hero shots, --ar 4:3 for presentation boards, --ar 2:3 for tall towers and sections. Style: --raw for cleaner geometry (--style raw on V7). Negative prompt: --no people, cars, text, watermark to strip clutter. Stylize: --stylize controls artistic license. Chaos (--chaos 0 to 50) widens the spread of options in the first grid, which is useful early in concept design.
It can imitate the look of plans, sections, and axonometric drawings, but they are not measured or buildable. The lines look convincing and are great for diagrams, competition boards, and concept communication, yet dimensions, structure, and code compliance are invented. Treat them as visual language, not documentation. For anything that needs to be accurate, draw it in CAD or BIM and use Midjourney only for the presentation atmosphere around it. If you already have a drawing and want a spatial view of it, a floor plan to 3D converter works from the real geometry instead of inventing one.
No. Midjourney starts at around 10 dollars per month for the basic plan, which is enough for most concept work. There is no permanent free tier. If your goal is photorealistic interior renders of a real room rather than exterior concept imagery, MeltFlex preserves the actual room from a photo instead of generating an invented space.
Midjourney invents space. Ask it to redesign a specific room you have already built or photographed and it will change the windows, move the walls, and ignore the dimensions. For exterior concept work that freedom is the point. For showing a client what their actual interior will look like after a redesign, it is a dealbreaker.
That is the gap MeltFlex fills, and it is the reverse of what Midjourney does: instead of inventing a building, it keeps yours. Here it is running on a real house exterior, redesigned from a single photo with the massing and openings left where they actually are:

The same building, restyled. Note the windows and roofline stay exactly where the photo put them, which is the thing a text prompt cannot promise.
Upload a photo of a real room and it preserves the exact walls, windows, and proportions while restyling the space with real furniture you can buy. Use Midjourney to win the concept, then use MeltFlex when the conversation moves indoors and accuracy starts to matter.
Copy any prompt above, paste it into Midjourney V8.1, and iterate on materials and light until it sings. Keep the six-part formula, lead with materials, and always test --raw against the default to see which serves the project. Lock one image as your --sref before you run a whole board. When the project moves from concept to a real interior, upload a room photo to MeltFlex and see an accurate redesign in seconds.
Try MeltFlex free: see a real room redesigned →
Other prompt libraries in this cluster:
When the concept has to become an accurate render:
Background and craft: