Midjourney for Architects: Concept Images for Client Meetings

Use Midjourney for mood, massing and materials, never as a design proposal. Label every image as an AI concept, show sets of alternatives instead of one hero render, keep dimensions and structure out of the conversation, and attach a plain disclaimer. Expect $10 to $60 a month; Stealth Mode, the privacy setting, starts on Pro.

The care is needed because a polished image reads as a promise. Clients see a render and hear "this is what you will get". Midjourney's documentation says it uses image prompts and references as inspiration, not to copy them exactly, so it cannot hold your footprint, levels or window positions. That is fine for atmosphere and a problem for anything that sounds like a design.

Follow me on Instagram@sagnikteaches

Preparing one client meeting this way takes about 90 minutes on Midjourney's website. Facts about plans, privacy and terms come from Midjourney's own pages in early November 2026, and they change, so check them before you subscribe. Every project detail is illustrative.

Connect on LinkedInSagnik Bhattacharya

What a concept image is for in a client meeting

Decide the job before you write a prompt. An image that is meant to start a conversation about light should never be asked to answer a question about layout.

Subscribe on YouTube@sagnikteaches

For anything in the right-hand column, show a drawing or a model you control. If the question is whether AI can produce floor plans at all, our tutorial on AI floor plan generation shows where it fails a real brief.

Plans, GPU time and what a month costs

Midjourney sells four subscription tiers, and its plan comparison page lists them in dollars. Annual billing, paid upfront, takes 20% off.

Extra Fast time costs $4 an hour on every plan. Midjourney's page on GPU speeds puts a standard-resolution (SD) image prompt at about 0.8 minutes of GPU time and an HD prompt at 1.3, while an Edit model prompt takes 1 minute in SD and 2.3 in HD, so reference-heavy sessions use more. The worked example later uses 22 jobs, which is 17.6 minutes at standard resolution or 28.6 if all were HD. Basic's 200 minutes covers about 11 such sessions a month at standard resolution or 7 in HD. Standard's 15 hours covers about 51 at standard, with Relax mode for unlimited images beyond that.

Two rules decide the plan. If any image comes from an identifiable client site, or you upload anything, you need Stealth Mode, so Pro. And Midjourney's plan page says a company making more than $1,000,000 USD in gross revenue a year must buy Pro or Mega, so a practice over that line has no cheaper option. A solo architect doing non-confidential mood work gets by on Standard. For other tools and what they charge, see our comparison of concept image tools by cost, and for the in-house or outsourced question, AI renders against outsourced visualisation.

Privacy first: public by default, and what you upload

Midjourney's terms say that, by default, your content is publicly viewable and remixable. That is the opposite of what a client assumes about their own house. Stealth Mode keeps images and videos private on the website, but it belongs to Pro and Mega only, and the terms promise "best efforts" not to publish what you make with it on, not a guarantee.

Two catches come from Midjourney's Stealth Mode page. Anything created in a public Discord channel is visible to others even with Stealth Mode on, so use the website's Create page, direct messages or a private server. And moving up to Pro does not make earlier public images private on its own: you select them on the website's page for organising your creations and change them yourself.

Uploads need their own rule. The terms make you responsible for having every right and permission needed for content you upload, ask you not to upload other people's private information, and give Midjourney a broad licence to reproduce and display what you input and what you create. A practice policy that fits in four lines:

1. No client drawings, surveys, property photos or names in prompts or uploads without written permission.
2. Describe building types and moods, not addresses.
3. Pro with Stealth Mode for anything tied to a real client project.
4. No other practice's published work as a reference for something you will present as your own idea.

For the wider copyright and client-data picture, our tutorial on copyright and client data risks of AI in architecture goes through the cases, and our practice AI policy checklist has clauses you can adapt.

Prompt patterns for massing

A massing prompt asks for simple volumes, a viewpoint and a medium that tells the viewer this is a study. The pattern is: what the object is, the volumes in plain words, viewpoint and light, the medium, then parameters. Midjourney's parameter page says parameters go at the end, after a space, and take no punctuation inside them.

Three massing prompts for the same illustrative house, each in a medium that reads as a study rather than a finished render:

A (model photograph): simple white card model of a two-storey house with a single-storey flat-roofed extension at the back, soft studio light, shallow depth of field, plain grey background, architectural massing study --ar 3:2 --stylize 50 --raw --no people, text, trees

B (sketch): loose ink and watercolour sketch of a pitched-roof brick house with a two-storey rear return, seen from the garden, overcast light, visible paper texture, concept sketch --ar 16:9 --stylize 150 --no people, text, cars

C (loft): pencil sketch elevation of a brick house in a row of similar houses with a timber-clad loft dormer, flat front view, no perspective, concept sketch --ar 4:3 --stylize 100 --no people, text, cars

Expect to fix a few things in the output (illustrative). Extra windows and doors appear that nobody asked for. A flat roof comes back pitched on one run in four. Roof pitch changes between runs of the same prompt, which is why each option should keep one base prompt and change one phrase at a time. A card model looks suspiciously perfect, so add "visible glue marks, plain card" if you want it to read as a study. And trees or people give a false sense of scale, which is why the prompts exclude them and you add context yourself in the deck.

A vague prompt against a specific one

The fastest way to mislead a client is a vague prompt. Compare "modern extension for a brick house, beautiful, realistic" with prompt A above. A vague prompt leaves Midjourney free to interpret, and its Stylize page says higher settings give more artistic images that can stray from the exact details of your prompt. Unspecific wording behaves in a similar way. You might get a glossy golden-hour glass box with a pool and palm trees (a typical failure pattern, not a recorded run): attractive, nothing like your house, and exactly the image a client will remember. Prompt A names the object, the volumes, the medium and what to leave out, so the result can only be a study of a house like yours. Specific wording also makes results repeatable, which you need when a client asks why option B looks different from option A.

Prompt patterns for mood and materials

Mood prompts describe light, materials and feeling, and they can be looser. Raise Stylize, and keep people out.

Interior mood: warm timber and lime plaster kitchen-dining room, low evening sun across a stone floor, large sliding doors to a garden, calm and uncluttered, interior mood image --ar 4:3 --stylize 250 --no people, text, logos

Materials board: close-up palette board on a plain table, charred timber cladding, hand-made brick, zinc standing seam, pale oak, overhead daylight --ar 3:2 --raw --no people, text

To show light, repeat one prompt and change only the light words: "low morning sun", "flat overcast light", "dusk with lit windows". The client then compares light, not form, because nothing else changed. For a consistent look across a whole deck, Midjourney's style reference feature matches the look of an image you supply, and our tutorial on keeping one look across a campaign covers the method.

Using your own sketch or model as a reference

Midjourney's Edit model, which replaced the older reference tools, can change an existing image from written instructions or generate from up to four reference images, and its docs say it works with versions 8.1 and 8.2 and with image prompts and style references. The practical use: photograph your own pencil sketch or foam model, or screenshot a plain grey massing view from your model (no client names in the file), and ask for a watercolour version. The composition will follow your reference. Window positions, proportions and roof pitch will drift, because the docs describe references as inspiration, not exact copies. Treat the result as a mood layer over your geometry, never as a drawing.

Show sets, and choose a style that reads as a sketch

Two habits do more than any disclaimer to stop clients misreading images. The first is to show sets: three or four images per option, from the same viewpoint, labelled A, B and C. One hero render invites a client to fall in love with a single picture; a set invites a comparison of ideas. The second is to pick a medium that signals "study". A card-model photograph, an ink-and-wash sketch or a pencil elevation tells the viewer this is thinking in progress. A photorealistic exterior at golden hour says the work is finished.

A deck order that works for a first meeting:

1. Purpose of the images, and the disclaimer.
2. The brief as you understood it, in five lines.
3. Option A, then B, then C: one slide each, with a sketch and a model photograph side by side.
4. A mood set of three or four images with materials and light.
5. A materials board.
6. Three questions for the client, such as which option felt closest and what they would change.
7. Next steps: survey, measured drawings, and a second meeting with real plans.

Crop out anything invented that could be mistaken for fact, such as a neighbouring building, a car or a sign, and rerun rather than retouch when a prompt keeps adding it.

The client disclaimer: wording and where to put it

A disclaimer only works if the client sees it before the pictures and again beside them. Use two layers: a short caption inside every image file, and a fuller paragraph on the second slide and in the email you send before the meeting. Sample wording follows, to adapt, not to treat as legal advice.

Caption on every image: AI-generated concept image for discussion of mood and massing only. Not to scale. Not a design proposal.

Paragraph for slide 2: These images were generated with an AI tool to explore atmosphere, materials and the general shape of a building. They are not drawings. They do not show our design, or dimensions, structure, services, planning compliance, cost or buildability, and features in them may not be possible on your site. We will develop the design in drawings and models checked against your brief, the site and the rules that apply. Please treat the images as prompts for conversation.

Email before the meeting: Hi [first name], for Thursday I have prepared a short set of concept images showing mood and massing options for the extension. They are AI-generated sketches for discussion, not design proposals, so please judge them on atmosphere rather than detail. I will explain what each one is for. [your name]

Words matter in the room as much as words on the slide. Avoid "this is what it will look like", "we can build that" and "something like this". Say "this is the mood we are exploring" and "this option has the same volume as A but a different roof". Say plainly that the images are AI-generated, because some clients and some professional bodies expect disclosure, so check what yours says. If your professional indemnity insurer or adviser wants specific wording, use theirs.

One house extension, 90 minutes of preparation

Here is the whole workflow on an illustrative job: a family wants a rear extension and a loft room in a two-storey brick house. The first meeting is on Thursday.

1. 20 minutes: brief and style. Write the brief in five lines, choose three massing ideas (A, a single-storey glazed box; B, a two-storey brick return; C, a loft dormer only), and fix the medium (card-model photograph plus ink-and-wash sketch) and the mood words (timber, brick, evening light).
2. 35 minutes: 22 jobs. Twelve massing runs, four per option, seven mood runs and three materials boards.
3. 15 minutes: select and label. Keep two images per option, rename them A-model, A-sketch and so on, and crop out invented context.
4. 20 minutes: the deck. Nine slides in the order given earlier, with the caption strip burned into each image.

The Fast time used is 22 jobs at 0.8 minutes, about 18 minutes, or about 29 if every job had been HD at 1.3. That is roughly 2% of Standard's 900 Fast minutes and about 9% of Basic's 200. The plan decision came from privacy, not capacity: the project is tied to a real family and address, so the work ran on Pro with Stealth Mode, a $60 month or $48 on annual billing.

Suppose the family prefers option B's two-storey return, likes the timber from the mood set and asks whether the large glazed doors in one image are possible. The answer is "that is a mood reference; we will test opening sizes against structure and cost in the next stage", and the next step is a measured survey and drawings, not more images. If you ask afterwards what they thought the pictures were for, the answer you want is "ideas".

What goes wrong in the room

These are the failures to rehearse for, each with a line you can use (all illustrative).

• The client fixates on an impossible feature. A glazed corner with no visible support, or a long cantilever, becomes "the one we want". Say: "That is a mood reference. The structure for it changes cost and consent, so we will test the idea with the engineer before we promise it."
• An image shows a recognisable neighbour or landmark. The client assumes it is real. Regenerate with a plainer prompt, or crop, and never show the image with the context in place.
• An image resembles a known building. Before any image goes on a website or in a public presentation, run a reverse image search on it. Keep client-only images in client-only decks.
• Different images show different house shapes. The client thinks the design is changing between slides. Keep one base prompt per option, change one phrase at a time, and label slides "mood only" or "massing only".
• Garbled signs and text. Add text to the exclusion list and crop what remains.
• The picture outlives the meeting. A client forwards one image to a builder or a friend without the slides. That is why the caption sits inside the image file, as a strip along the bottom, and not only in the deck.

A concept image log for the project file

Midjourney keeps your creations on its site, but it will not tell you which image went to which client with which caption. A one-row-per-image log does. It also answers the awkward question six months later of what you actually showed.

Store the full prompt text, not a summary, because small phrase changes explain big visual differences. Note the version, since Midjourney's default has moved several times (V8.2 became the default on 24 July 2026 and replaced earlier reference tools with the Edit model). And record whether the client gave permission for any photo or drawing you uploaded.

If you only change one thing after reading this, make it the second layer of disclosure: the caption inside the image. Everything else here improves the pictures. That one line protects you when an image travels without you. If you need more than mood, the floor plan tutorial linked earlier shows where AI stops being useful. Try one meeting this way before you commit to a plan.

Midjourney questions architects ask before using it with clients

Can Midjourney read my architectural plans or model?

Not as drawings. Midjourney's own documentation says it uses image prompts and references as inspiration, not to copy them exactly, so expect the composition to follow your reference while proportions, window positions and roof forms drift. Use it for mood, and render from your own model whenever the geometry must be right.

Can I use Midjourney images commercially in client work?

Its plan page says that once you have subscribed you can use your images in just about any way you want, but a company making more than $1,000,000 USD in gross revenue a year must be on Pro or Mega. The terms also tell you to consult your own lawyer about ownership, so check the current wording before relying on this.

Are my Midjourney images private?

By default content is publicly viewable and remixable on the Midjourney website. Stealth Mode, available on Pro and Mega, keeps images off the public site, but anything made in a public Discord channel stays visible regardless. The terms promise best efforts, not a guarantee, so keep client-identifying detail out of prompts.

How many images will the cheapest plan make?

Midjourney's GPU speed page puts a standard-resolution image prompt at about 0.8 minutes of Fast time and an HD prompt at 1.3, and Basic includes 200 minutes. That is roughly 250 standard prompts or 150 HD ones a month, and each prompt returns a set of four images. Edit model prompts use more, so check your usage after the first session.

Further reads

• How to Use AI in Architecture Design: Concept Options for Small Practices — More ways to generate concept options for a small practice.
• Best AI Rendering Tools for Small Architecture Practices (2026) — Rendering tools compared when you need more than mood.
• AI for Architects: Draft Design Schematics From a Client Brief — Turn a client brief into draft schematics, with limits.
• Is Midjourney Worth It for Small Business Marketing Images? — Whether the subscription pays for itself on marketing images.
• AI Room Visualisation From Client Photos: A Designer's Workflow — Start from the client's own room photo instead.
• How Interior Designers Use AI for Mood Boards and Presentations — Mood boards and presentations as interior designers do them.
• AI Marketing for Architects: Case Studies, Website Copy and LinkedIn — Use finished-project images and copy in practice marketing.
• AI Tools for Architects 2026: A Toolkit From Brief to Handover — Which AI tool to use at concept, design, tender, site and handover, what each costs on its own pricing page, and three stacks for a five-person practice.
• AI Tools for Landscape Architects: Planting Plans, Renders, Proposals — A landscape practice's workflow: plant lists with a four-point check, schedule sums, honestly labelled concept renders and staged fee proposals.
• Architect Website: Project Descriptions and Service Pages Written With AI — A copy workflow for architects: project descriptions from verified facts, service pages by project type, honest area pages and an FAQ on fees and process.
• Top AI Tools for Architects: Eight Compared by Job and Price — Eight AI tools matched to five practice jobs, with USD entry prices, plan limits and a monthly cost for a five-person practice.
• AI for Architects and Interior Designers: Who Does What on a Joint Job — Who should own which AI task when an architect and interior designer share a project: brief, mood boards, schedules, client updates and rights.
• Can AI Concept Images Help Win a Home Renovation Proposal? — How a renovation firm can use ChatGPT or Gemini concept images from a site photo in a proposal, with a not-a-design disclaimer and scope control.
• Midjourney for Graphic Designers: Client Work, Licences and Limits — What Midjourney's terms let a designer do with client work, how the reference tools behave, and the disclosure and logo limits to set.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Midjourney documentation (comparing plans, Stealth Mode, Terms of Service, version history, GPU speeds, parameter list, Image Prompts, Style Reference, Edit Model, Stylize, the No parameter), read in early November 2026. Project details and costs in the worked example are illustrative.