Interior designers use image generators for early concept boards that set a direction, then replace every AI image with a real, purchasable product before the client sees a selection board. AI also turns the brief into a colour and material vocabulary, drafts the design narrative and helps build the deck. Concept boards can be imagined; selection boards must be real.
That split is the whole method, and most problems come from blurring it. A client who falls for an AI-generated sofa, pendant or tiled floor will ask where to buy it, and the honest answer is that it doesn't exist. The second problem is sameness: AI images drift towards a polished average of every interior photo online, so without the brief's specifics every scheme ends up in the same beige, arched, boucle-chaired room. The steps below are designed around both.
Concept boards and selection boards do different jobs
| Concept board | Selection board | |
|---|---|---|
| Purpose | Agree a direction: mood, palette, era, feel | Agree specific products, finishes and budget |
| When | After the brief, before any sourcing | After the concept is approved |
| AI's role | Generate images, draft the narrative | Find real matches, capture product data, lay out |
| Images | AI-generated or precedent photos, labelled | Real product images only |
| Client takes away | "Yes, that feeling, but warmer" | "Order that sofa in that fabric" |
Turn the brief into a visual vocabulary first
Before generating a single image, ask AI to turn the brief into words you can prompt with. Paste the client questionnaire, your meeting notes and a description of the images they've saved (or the images themselves, in a tool that accepts them):
From this client brief and their saved images, write a visual
vocabulary for the scheme:
- 5 words for the mood, using the client's own words where
possible
- A palette: 4-6 colours described in plain terms
- Materials and finishes they're drawn to
- Period or style references, if any
- A "not this" list: anything they said they dislike
- Contradictions I should resolve with the client
Quote the brief for each point.
Brief, notes, image descriptions: [paste]An illustrative output for a couple redoing a living room and kitchen in a narrow 1900s townhouse:
Mood: calm, "lived-in not showroom", warm, uncluttered, bookish
Palette: warm off-white, mid oak, deep green, soft terracotta,
aged brass
Materials: oak, linen, honed stone, brass, wool rugs
References: "old but not fussy"; original cornicing to keep
Not this: grey, high gloss, "anything that looks like a hotel"
Contradictions: wants "uncluttered" but also "all our books
on show" (roughly 600 books)One fix before prompting: "soft terracotta" came from the model reading "warm" as a colour. Going back to the notes, the client had said "warm" about the light and the oak, and at the meeting had said they found orange tones "a bit much". Terracotta came off the palette. The contradiction line was genuinely useful: 600 books and "uncluttered" is a design problem to solve with joinery, and it's better raised at concept stage than discovered at installation.
Commercial briefs pull the vocabulary towards use. For a small café refit, add a "wear and cleaning" line to the prompt: tables wiped twenty times a day, chairs dragged across the floor, a counter front that takes every spill. Leave it out and the concept images will show velvet banquettes and unsealed oak tabletops, and the owner will fall for a look that's stained within a month. Put "commercial use: wipe-clean, hard-wearing" in the vocabulary and carry it into every image prompt.
Concept images that don't all look the same
Generic prompts produce generic rooms. Build each prompt from the vocabulary, the actual room and its constraints:
Narrow townhouse living room, about 3.5m wide, single bay
window facing north, original ornate cornice, pale oak floor.
Built-in oak shelving floor to ceiling on one wall, filled with
books. Deep green linen sofa, aged brass reading lamp, wool rug.
Soft, cool north light with warm lamp light. Calm, lived-in,
uncluttered. Eye-level photograph from the doorway.
Avoid: grey, gloss, arches, boucle, hotel-style styling.Tools that suit this stage, at USD list prices checked in September 2026:
- Midjourney, from $10 to $120 a month. Strong on mood and light. On the cheaper plans your images appear in its public gallery; only the Pro ($60) and Mega plans include Stealth mode. For client homes, that's worth paying for. Its default model has been V8.2 since July 2026, and its Edit model replaced older tools such as Retexture, so guides that tell you to use those are out of date.
- Adobe Firefly, from $9.99 a month (Standard, 2,000 credits). Every plan includes Firefly Boards, a canvas for gathering reference images and generating new ones side by side, which suits mood-board work well.
- The image tools in ChatGPT or Gemini, if you already subscribe. Good for quick variations: "same room, walls in warm off-white instead".
Generate eight to twelve images and choose three or four. Reject any that break the room's reality. From an illustrative batch of twelve, typical rejects: two images added a fireplace on a wall where the chimney breast was removed years ago; one turned the single bay into two sash windows; one tripled the room's width. None would have been obvious to the client, which is exactly why you check. If you want AI to work from photographs of the client's actual room instead, that's a different workflow with its own pitfalls, covered in AI room visualisation from client photos.
Kitchens need a stricter eye than living rooms, because the images imply a layout. Suppose one image in the kitchen round puts the hob in a central island with nothing above it, and the clients' first question at the presentation is whether they can have "the island in picture 3". In a narrow townhouse kitchen, an island hob means a ceiling or downdraft extractor, ducting, and possibly less worktop than they have now. Don't hide the image; say at the start that concept images show mood and materials, not layout, and bring the layout options on the plan to the same meeting. Better still, prompt the kitchen from the plan you already have: "sink under the window, hob on the side wall, no island".
Label the board as well as the images, because boards travel. A client who forwards the concept board to their builder "for a rough idea" can come back with a joinery price for shelving drawn with no visible supports and no thought for what 600 books weigh. A footer line on every concept board heads that off:
Concept imagery, partly AI-generated. Shows mood, palette and
materials only. Products, sizes and construction are confirmed
on the selection board and drawings.It takes a minute to add to your board template, and it gives the builder, and you, something to point to.
Turning client feedback into the next round
Concept feedback often arrives as a long, warm, slightly contradictory email, sometimes from two people who don't quite agree. AI is good at turning it into a change list you can act on and confirm. An illustrative message after the first concept presentation:
"We loved it! The green is gorgeous, though maybe a bit dark for the living room? [Partner] thinks the brass is too shiny but I love it. Could we see something with more pattern? Not sure about the rug. The shelves are perfect, exactly what we wanted."
Ask for three lists: agreed, changes requested, and disagreements to resolve. The result, after a quick check against the email:
Agreed: floor-to-ceiling oak shelving; overall mood
Changes: try a lighter green on the sofa; explore more
pattern (where? ask); alternative rug options
Disagreements: brass finish (one partner likes it shiny, one
doesn't) -> show aged/brushed brass as a middle optionThe one thing to add by hand: "more pattern" needs a follow-up question before any new images are generated, because it could mean cushions, curtains, a rug or wallpaper, and guessing wastes a round. Sending the change list back to the clients for a quick "yes, that's right" also gives you a record of what they asked for, which helps when the third round of changes starts to look like a new scheme.
From concept to real products
Once the client approves the direction, every AI element needs a real counterpart. This is where the time goes, and where AI helps in a different way: finding and capturing products rather than inventing them.
- Visual search. A reverse image search such as Google Lens on a cropped AI image of a lamp or chair will often surface real products with a similar look. Treat results as leads; check dimensions and availability on the maker's site.
- Clippers in design software. Houzz Pro's AI-powered clipper captures product images, pricing and specifications from most supplier websites into your product library, and its mood boards and visual proposals let clients review and approve selections. Programa's clipper saves products from supplier sites straight into a project's schedule. Both keep the real data (price, supplier, lead time) attached to the image, which a folder of screenshots never does.
- AI for comparisons. Paste three candidate sofas' specifications and ask for a table of dimensions, seat depth, fabric options, lead time and price. Check the table against the pages.
The comparison step needs one extra instruction. Pasting three sofa pages and asking for that table returned, in an illustrative run, a tidy grid with a seat depth for all three. Only two pages gave one; the third figure, 55cm, was the model filling a gap with a plausible number. Seat depth decides whether a tall client is comfortable, so it's the worst figure to guess. Add "write 'not listed' for any value the page doesn't state", and ask the supplier for anything missing before it reaches a selection board.
Visual search has its own trap. Crop an AI lamp that happens to resemble a well-known designer piece and the results fill with near-identical replicas at a quarter of the price, from sellers who name no maker and give vague lead times. Replicas of protected designs can raise design-rights and quality problems, and a client who later learns their "designer" lamp is a copy won't thank you for it. Put the original maker's version on the selection board, or choose a genuinely different piece that gives the same effect, and note which you've done.
An illustrative swap sheet for the living room:
| Concept image element | Real product found | Price | Lead time | Note |
|---|---|---|---|---|
| Deep green linen sofa | Three-seater in green washed linen, 210cm | $3,400 | 10 weeks | AI sofa looked 240cm; real one fits the alcove |
| Aged brass reading lamp | Floor lamp, antique brass, adjustable arm | $620 | In stock | Close match |
| Floor-to-ceiling oak shelving | Bespoke joinery, local maker | Quote pending | 8-12 weeks | AI image had no visible supports; joiner needs to advise |
| Wool rug, muted pattern | Handwoven wool, 200 x 300cm | $1,150 | 4 weeks | Pattern less busy than concept |
The notes column is what you'll talk the client through: where reality differs from the concept, and why that's fine.
Building the presentation deck
A structure that works for most residential schemes: the brief in the client's words; the concept board; the plan and any layout options; a selection board per room; a budget summary; next steps and timescale. AI can draft the narrative for each section and, in PowerPoint with Copilot or in Canva, help lay it out from your content.
The budget summary is where AI helps with totals and needs the hardest check. From the swap sheet above, the known items come to $5,170: the sofa at $3,400, the lamp at $620 and the rug at $1,150. The joinery quote is still pending, so the slide shows it as the range from the joiner's first estimate, never a number AI proposes, plus a contingency line (say 10% of known items, $517) for price changes over the lead times. Ask AI to total the swap sheet, then re-add it yourself or in a spreadsheet. A budget slide that doesn't add up is the one slide every client remembers.
The narrative needs a firm hand. Ask for a short concept statement from the vocabulary and you'll often get something like this (illustrative first draft):
First draft: "This scheme creates a timeless haven that effortlessly marries form and function, celebrating the heritage of the home while embracing contemporary living through a curated palette of natural materials."
It could describe any room in any house. Rewrite with the client's words and one concrete idea:
Rewritten: "You asked for a room that feels lived-in, not a showroom, with all your books on show. The scheme puts them on floor-to-ceiling oak shelves along the long wall, so the books become the room's main feature and the rest can stay calm: warm off-white walls, a deep green linen sofa, brass lamp light for the dark north-facing evenings."
Adding "use the client's own words, mention one specific design move, no more than 70 words, avoid the words timeless, curated and haven" to the prompt gets you most of the way there on the first try. The same discipline applies to any AI-written client document; AI for freelance designers covers concept work and rights from the graphic-design side.
A two-person studio's living room and kitchen project
For a two-person studio on a living room and kitchen scheme, the arithmetic runs roughly like this. Before using AI, the concept board takes about four hours (searching image libraries and collaging), sourcing about six hours, and the presentation deck about five: fifteen hours of largely unbillable work, or billed at a rate the client notices.
With the method above, the vocabulary step takes twenty minutes, concept images and the board about ninety minutes, sourcing with a clipper around four hours (it still takes judgement to choose the right sofa), and the deck about two hours with AI drafting the narrative and layout. Roughly eight hours in total: seven saved per project. For a studio running three such projects a month, that's around 20 hours a month, which can go into more projects or simply shorter evenings.
On cost, Firefly Standard at $9.99 a month or Midjourney Pro at $60 a month for private generation, plus whichever assistant and design platform the studio already uses. Check Houzz Pro's and Programa's current pricing on their own sites; both quote by plan and seat. The admin around the project (quotes, supplier chasing, client updates) is a separate saving, covered in AI for interior design studio admin.
Where AI boards create client problems
- Screens lie about colour. An AI image of "deep green" on a phone looks nothing like the paint on a north-facing wall. Always follow concept colour with physical samples in the room.
- Scale is invented. AI rooms are often larger and furniture smaller than reality. Put key dimensions on the selection board so the client isn't comparing a 210cm sofa to an imaginary 240cm one.
- Impossible details. Shelving with no supports, lights with no visible fitting, rugs that curve up walls. Check every image before it leaves the studio; checking AI-generated images for errors has a quick routine.
- Privacy of client homes. Uploading photos of a client's home to a tool whose images are public, or to a personal account, exposes more than you'd expect: layouts, valuables, entry points. Use private settings and business accounts.
- Portfolio honesty. Posting AI concept boards as if they were completed work misleads future clients. Label them as concepts; see whether you can legally use AI images in marketing before using them publicly.
Questions designers ask about AI boards
Should I tell clients a mood board uses AI images?
Yes. A small label such as "concept imagery, AI-generated" on each AI image prevents a client asking where to buy a sofa that doesn't exist. Most clients are relaxed about AI at concept stage when it's explained as a way of showing direction quickly. The trust problem comes when they find out later.
Can I use AI mood board images on my website or social media?
Check the tool's terms for commercial use first, and remember that some cheaper plans make your images publicly visible. Label them as AI-generated concept imagery rather than presenting them as completed projects, which would mislead potential clients. Photographs of finished rooms are still the strongest portfolio material.
Do AI tools understand specific paint colours?
Not reliably. Asking for a named paint colour gets you something in the right family, rendered with invented lighting on an uncalibrated screen. Use AI images for mood, then show real paint samples, ideally painted onto large boards and viewed in the client's own room at different times of day.
Further reads
- Is Midjourney Worth It for Small Business Marketing Images? — Whether a Midjourney subscription pays its way.
- Best AI Rendering Tools for Small Architecture Practices (2026) — Rendering tools architects use, several of which suit interiors.
- How to Label AI-Generated Images and Video on Social Media — Labelling AI imagery when you share it publicly.
- How to Write Business Proposals Faster With AI — Turn the approved scheme into a written proposal.
- Brand Guidelines for AI Images: Colours, Style and Limits — Keep your studio's own style consistent in AI images.
- Claude vs ChatGPT for Business Writing, Proposals and Reports — Which assistant writes better design narratives.
- AI Virtual Staging for Property Listings: Costs and Disclosure — Per-photo and per-listing costs of AI virtual staging, the edits a label can't excuse, label wording to copy, and a pre-publish check for every image.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Adobe Firefly plans page; Houzz Pro mood board and clipper feature pages; Programa product information; Midjourney plan details. Checked September 2026.