Most small studios can cover concept images for about $40 to $150 a month: Midjourney (Standard about $30; its terms, as widely quoted, put companies above $1 million in annual revenue on Pro, about $60) or Adobe Firefly (from $9.99) for general concepts, plus Veras or mnml.ai for model and sketch renders. Free plans are usually non-commercial.
The monthly fee is the smaller decision, so check the licence before a client sees anything. A studio's real exposure sits in three terms: whether a plan can be used commercially at all, whether the images are public by default, and what the vendor will stand behind if a client's use is challenged. Your own time costs more than any subscription here, so choose on control and rights first and on price second.
What each concept-image route costs a month
These are list prices from each vendor's own pages in late October 2026. Midjourney's pages did not load for me, so treat its figures as search-sourced and confirm them on the plans page before buying.
| Route | Plans and list prices | What you get | Watch for |
|---|---|---|---|
| Midjourney | Reported as Basic $10, Standard $30, Pro $60, Mega $120 a month, with about 20% off on annual billing | Text and image-prompted concepts; Stealth Mode reported on Pro and Mega | Images reported public by default; revenue threshold of $1 million |
| Adobe Firefly | Free daily generations; Standard $9.99 (2,000 credits), Pro $19.99 (4,000), Pro Plus $49.99 (10,000), Premium $199.99 (50,000) | Paid plans include unlimited standard image generations with Adobe's own models; partner models spend credits | Commercial-safety wording covers Adobe's models, not partner models |
| ChatGPT images | Free is limited; Go about $8, Plus about $20, Pro from $100 a month at list price | Generate or edit images in conversation, including from a client's uploaded photo; limits vary by plan | Consumer plans have a training switch; seats need a business plan |
| Veras (EvolveLAB) | Named licence $59 a month, or $29 billed yearly ($348); floating licence $51 billed annually ($612); student $149 a year | Renders from SketchUp, Revit, Rhino, Archicad, Vectorworks and Forma views; 30 renders and a 15-day trial | Commercial terms for named and floating licences are not stated on the page |
| mnml.ai | Lite $29 ($290 yearly), Plus $49 ($490), Studio $119 ($1,190); a one-off $149 pack of 10,000 credits valid 12 months | About 100, 300 or 1,000 fast renders a month; 8K upscaling; PNG and DXF export; 100 free credits at sign-up | Free-plan assets are non-commercial and watermarked |
| Stable Diffusion run locally | Free licence below $1 million in annual revenue | Full control and no per-image fee | Needs a capable computer and a person to run it; commercial users must register |
Hidden costs worth pricing in: annual billing is a 12-month commitment (Veras's named licence is $708 over a year on monthly billing and $348 on yearly billing), credit top-ups when a plan runs dry, retouching time, and rework if a licence turns out not to cover the use. For the wider tool landscape, the rendering tools for small practices compares features, and Midjourney for architects covers prompts and client meetings.
Commercial licences: the thresholds and free-plan traps
Four terms decide whether a plan covers client work.
- Revenue thresholds. As Midjourney's terms are quoted in search results (its own pages would not load for me), a company with more than $1 million in annual gross revenue must be on Pro or Mega to own its assets. Stability AI's community licence draws the same line: free commercial use below $1 million in annual revenue, and registration with Stability for commercial users. Above it, the community licence ends and you must request a licence, which Stability may grant at its discretion. The figure is revenue per year, so a studio at $900,000 this year and $1.2 million next should plan for the higher tier a year early.
- Free plans and student licences. The mnml.ai terms say assets created on the free plan may be used only for personal, non-commercial purposes and carry a watermark, while paid plans cover personal or commercial use. Veras labels its student licence non-commercial.
- Whose models. Adobe's Firefly plans page says outputs from Adobe Firefly models are safe for commercial use, and that for partner models you are responsible for deciding whether the content is appropriate for your project. Firefly now offers partner models from other companies, so check which model made an image.
- Silence. Veras's pricing page says nothing about commercial use or ownership for its named and floating licences. Silence is not a licence, so ask for the terms in writing before using renders in a client deliverable.
For one tool's rules in depth, see which Midjourney plan a business needs for commercial use.
Who owns the image, who can see it, who covers a claim
Ownership is only the first of the client-use questions. A studio also needs to know whether the image sits in a public gallery, whether uploads such as a client's photo or plan are kept or used for training, and whether the vendor offers any protection if a third party objects.
| Route | Ownership wording | Visibility | Your uploads | Protection if challenged |
|---|---|---|---|---|
| Midjourney | Reported to depend on being on a qualifying paid plan (see the threshold above) | Reported as public and remixable by default, with Stealth Mode on Pro and Mega described as best efforts; confirm on its terms page | Not covered in the sources I could read | Not stated |
| Adobe Firefly | Outputs of Adobe's own models described as safe for commercial use | Not stated on the plans page | Not stated on the plans page | IP indemnity is limited to enterprise customers; Adobe's guidelines forbid removing Content Credentials |
| ChatGPT | Terms assign OpenAI's rights in the output to you, "if any" | Not a public gallery; check your sharing settings | Consumer plans have a training switch in privacy settings; business plans do not train by default | Output indemnity covers API and Enterprise customers, not ChatGPT Business |
| Veras | Not stated | Not stated | Not stated | Not stated |
| mnml.ai | "Any assets created using the Service are owned by you" | Private by default; public only if you publish a render, and published renders may be featured on its pages | No training on uploaded content or generated assets | None offered; its terms have you indemnify mnml.ai against claims arising from your use |
| Stable Diffusion, local | You own outputs "to the extent permitted by applicable law" | Stays on your machine | Stays on your machine | None |
Two cautions on reading that table. "You own the output" is a contract statement between you and the vendor. It does not mean the image is protected by copyright, and it does not protect you from a third party's rights, such as a recognisable building, a trade mark or a living architect's distinctive design reproduced by a style prompt. And "not stated" in a cell means I found nothing on the pages I read, which is a prompt to ask, not evidence of a problem. Copyright and client-data risks of AI in architecture practices goes further on both.
A ten-minute licence check, filled in for two render tools
Before a render goes into a proposal, run the vendor questions listed further down against the plan you actually pay for. Here is the check filled in for two render tools a three-person studio might shortlist, using only what each vendor's pages said when I read them in October 2026.
| Question | mnml.ai Plus, $49 a month | Veras named licence, $59 a month or $29 billed yearly |
|---|---|---|
| Commercial use on this plan? | Yes: the terms allow paid-plan assets for "any personal or commercial purpose", and the pricing page lists a commercial licence on every plan | Not stated for named or floating licences; the student licence is non-commercial |
| Who owns the output? | You do, per the terms | Not stated |
| Public by default? | No: private unless you publish a render, and published renders may be featured | Not stated |
| Uploads used for training? | No, per the terms | Not stated |
| Cover if a third party objects? | None; you indemnify mnml.ai | Not stated |
| Watermark? | Only on free-plan assets | Not stated |
| Verdict | Usable in a client pitch on a paid plan | Ask for the commercial terms in writing before a client sees a render |
The Veras column is not a criticism of the product; it is a list of questions to send. Ask them, file the reply with the project, and the check is complete. The mnml.ai column carries the other lesson: even a clear licence leaves the indemnity running from you to the vendor, so a render that copies a recognisable building or another designer's signature detail is your risk, not theirs. Repeat the check whenever a plan changes, because terms are revised more often than prices.
Three studios, three monthly bills
An illustration with list prices. Three invented studios, each pricing only what its licence and workload require.
| Studio | Needs | Stack | Monthly cost |
|---|---|---|---|
| Solo interior designer, revenue well under $1 million | Mood boards and room concepts from client photos | Firefly Standard $9.99 plus Midjourney Standard $30 | $39.99, or $479.88 a year |
| Three-person architecture studio, revenue under $1 million | Massing and facade options from SketchUp or Revit views, plus mood images | Midjourney Standard $30 plus two Veras named licences at $29 each, billed yearly | $88, or $1,056 a year ($148 a month if Veras is billed monthly) |
| Twelve-person practice, revenue over $1 million | Concepts for client presentations and competitions | Midjourney Pro $60 plus one Veras floating licence $51 plus Firefly Standard $9.99 | $120.99, or $1,451.88 a year |
The working for the middle row: $30 + 2 x $29 = $88 a month. The $1,056 a year is $360 for Midjourney plus $696 for two yearly Veras licences. On monthly billing the two Veras seats are 2 x $59 = $118, which with Midjourney makes $148. For the twelve-person practice, the threshold does the choosing: Pro at about $60 is not a luxury, it is the plan the terms are quoted as requiring, and the reported Stealth Mode comes with it. A studio could swap Veras for mnml.ai Plus at $49 if it works from sketches and photos rather than a model.
Which route fits which job
A single studio rarely needs all six routes. Match the route to the job, and the bill takes care of itself.
| Job | Route that fits | Why |
|---|---|---|
| Early mood and material direction | Midjourney or Firefly | Fast variations, and low risk because nothing gets built from them |
| Changing a client's real room | ChatGPT images or Firefly, with the client's permission | Starts from the client's photo, so the upload and training terms matter most |
| Massing and facade options from a model | Veras, or the AI renderer inside your modelling software if it has one | Works from a view of your model, so the geometry you drew guides the render; check how closely on your own project |
| Sketch to render with a drawing-friendly export | mnml.ai | PNG and DXF export, and a commercial licence on every paid plan |
| No budget and a technical person on staff | Stable Diffusion run locally | Free below the revenue line, with the support burden on you |
Before buying any of these, check whether your modelling software already includes an AI renderer on the plan you pay for. Several vendors have added one, and an included tool needs the same eight licence questions listed below but no new subscription.
Running Stable Diffusion yourself: a free licence with real costs
The Stability AI community licence lets organisations under $1 million in annual revenue use its models commercially at no charge, says you own the outputs to the extent the law permits, and requires commercial users to register with Stability. The licence is free; the work is not. Someone has to install the software, keep it updated, choose models, and write down which model produced which image. If nobody on the team enjoys that, a $30 plan is cheaper than the hour a month it takes to maintain a local setup. If someone does, it is the one route where client photos never leave your machine.
The cost nobody lists: your own time
Take an assumption and run it. A concept board takes about 45 minutes of prompting, choosing and tidying. At $60 an hour of studio time that is $45 a board; eight boards a month is $360 of time against a $30 plan, so the time costs 12 times the subscription. That reverses the usual question. A plan or tool that follows instructions better and saves 10 minutes a board is worth $10 a board, or $80 a month at eight boards, which is more than the gap between most plans in the table. Pay for control and clear rights, not for the cheapest credit.
From a client's room photo to a safe concept image
Editing a client's own photo is the most common request, and the one with the most ways to go wrong. Ask the client first, and check whether your plan trains on uploads before their photo goes in. This prompt keeps the structure fixed and changes only the surfaces.
Use the attached photo of the living room as the base image.
Keep the walls, windows, door positions and ceiling height exactly as they are.
Show the room with a limed-oak floor, warm white walls, a deep green built-in
alcove either side of the fireplace and a linen sofa under the window.
Do not move the radiator or add furniture that blocks the door.
Soft daylight, realistic photographic style, no text, no people.An illustrative result: the image looks convincing, but the window is wider than the real one, the radiator has disappeared, and the alcoves read about 60 cm deep where the real recesses are 35 cm. None of that is a failure of the tool; it is what image models do with a loose instruction. What I would do next:
- Lay the output over the original photo and note every change you did not ask for.
- Re-run with the missed constraint named, "keep the radiator under the window", and again for the alcove depth.
- Caption the final image honestly. Before: "Proposed living room." After: "Concept image, AI-assisted: mood, colour and materials only. Window, radiator and alcove sizes are approximate; see drawing 12 for dimensions."
- Keep the prompt, the tool and the plan with the project file, so you can show what produced the image.
For the interior-design side of this workflow, the guide to AI mock-ups and client-ready images for a studio has more examples.
From a pencil elevation to a concept render
For an architect the input is usually a sketch or a model view, not a photograph. Tools such as Veras and mnml.ai are built around that input; a general image tool can do it too if the prompt pins down what must not change.
Use the attached pencil elevation as the structure. Render a two-storey rear
extension in mixed red brick with a zinc standing-seam roof, overcast daylight
and a neutral garden setting. Keep the window and door positions and the roof
pitch as drawn. No balconies, no extra storeys, no people, no text.A typical illustrative result gets the brick colour and light right, then adds a Juliet balcony nobody drew, flattens the roof pitch and moves one window. The checks are the same as for the room photo: overlay the render on the elevation, list every unrequested change, and re-run with each one named. Then caption it "Concept image, AI-assisted; the elevation drawing governs." The render persuades a client to feel something about the building. The drawing is still the only document that says what the building is.
Two records that protect the studio
A clause in your terms and a log in the project file cost an afternoon and answer most later disputes. This clause is a starting point for a solicitor to adapt, not legal advice.
Concept images. Concept images in our proposals may be created with the help of AI tools. They illustrate mood, massing and materials only. They are not design drawings, do not show a buildable or approved design, and do not form part of the contract documents. Dimensions, structure, services and finishes appear only on the issued drawings and specification. We will tell you which tools we used on request, and we do not upload your photographs, plans or personal details to any tool that publishes images publicly.
The log is one row per image. Here is one, filled in.
| Field | Entry |
|---|---|
| Date and project | 12 October, kitchen extension |
| Tool and plan | Midjourney Standard; visibility checked: images public on this plan |
| Inputs | Studio sketch only; no client photographs |
| Licence checked | Yes: studio revenue under the $1 million line, reviewed 1 October |
| Caption used | Concept image, AI-assisted; mood and materials only |
| Used in | Proposal, page 4 |
If your practice has a written AI policy, put both items in it; the AI policy checklist for architects shows where they fit.
A free-plan render in a client pitch, and three other slips
- The free-plan render. A studio pitches a kitchen extension with renders made from free sign-up credits on a render service. The terms for free-plan assets allow personal, non-commercial use and add a watermark. The client spots the faint mark, and the studio has to pay and regenerate, or drop the images. This is a plausible slip, not a reported one, and it costs a plan fee plus an awkward call.
- The public gallery. A designer generates variations from a client's new-build photos in a tool where creations are public and remixable by default. The client's unbuilt house is visible to strangers before the client has seen the finished boards. Check the visibility setting before the first upload. Midjourney's Stealth Mode is reported to be a best-efforts promise on Pro and Mega only, so read its current terms before relying on it.
- The stripped label. A practice exports Firefly images through an editor that removes metadata. Adobe's guidelines say you must not remove, alter or disable Content Credentials, so the habit breaks the terms of use.
- The image that became the design. A client points at a concept image and asks for "that staircase". It was never drawn, and does not meet the rules for a staircase. The caption and the clause above exist for this moment.
Eight questions to put to a vendor in writing
Send these before a tool becomes part of client work, and keep the replies with the log.
- Is my exact plan licensed for commercial use, and since when?
- Does a revenue or seat limit change which plan I need?
- Who owns the output, and does the vendor keep any right to display or reuse it?
- Are my images public by default, and what does the private mode actually promise?
- Are uploads such as client photos and plans kept, and are they used to train models?
- Is there an indemnity if a third party objects, and which plans does it cover?
- What labels or metadata does the tool attach, and may I remove them?
- What happens to my images and my licence if I cancel, or the vendor closes the product?
The last one is not theoretical. OpenAI closed its Sora app on 26 April 2026, and image and video tools change terms and plans often. Keep your own copies of finished images and your prompts, and treat any single vendor as replaceable.
Concept-image questions from design studios
Can I copyright concept images made with AI?
Often not in full. Several markets have said that images produced purely by a machine are not protected, and protection for AI-assisted work depends on how much human creative input went in, which differs by place. A tool's terms saying you own the output do not create copyright. Ask a solicitor before promising a client exclusivity, and rely on your drawings and designs for protection.
Should I tell clients that images are AI-generated?
Yes, in a short caption and in your terms of engagement. It protects you if a client later treats a concept image as a promise, and labelling rules for AI-generated images are tightening in some markets. A line such as 'concept image, AI-assisted, mood and materials only' on each image costs nothing and heads off most misunderstandings.
Is a free plan enough for pitch images?
Usually not for anything a client sees. The mnml.ai terms limit free-plan assets to personal, non-commercial use with a watermark, Veras's student licence is non-commercial, and other free tiers carry similar limits. Use free credits to test a tool, then pay for the plan before generating anything that goes into a proposal.
What if my practice crosses $1 million in revenue?
Midjourney's terms, as widely quoted, require Pro or Mega for companies above $1 million in annual gross revenue, so confirm on its terms page. Stability AI's free community licence stops at the same figure. If you expect to cross it within a year, move up early: at the reported $30 and $60 prices, Pro costs about $360 a year more than Standard, which is small against the risk of images you cannot rely on.
Further reads
- AI Room Visualisation From Client Photos: A Designer's Workflow — A designer's workflow for visualising rooms from client photos.
- AI Floor Plan Generation for Architects: Where It Fails a Real Brief — Where AI floor plans fail a real brief.
- Can AI Design a Restaurant? Layouts, Concept Images and Limits — AI can build a restaurant mood board and a design brief in a weekend. A pre-designer checklist covers what its layouts miss on fire, access and kitchen flow.
- Interior Designer Software Compared: Where AI Add-Ons Earn Their Fee — Six interior design tools priced per month, the AI credits worth buying, and a 14-day trial you can run on one finished project.
- 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.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Midjourney plans and terms as summarised in search results (its own pages blocked automated readers); Adobe Firefly plans page and generative AI user guidelines; ChatGPT pricing page; Veras (EvolveLAB) product page; mnml.ai pricing and terms pages; Stability AI licence and community licence agreement pages.