Use AI for the look and the brief, not the building. Image tools can build a mood board of materials, colour and lighting in an evening, and a text assistant can turn your ideas into a one-page brief. Treat any AI floor plan as a sketch: exits, access, capacity and kitchen rules need a professional.
The usual failure is a lovely render that seats twice as many people as the room can hold, has no route for dirty dishes and shows no door for deliveries. Image generators draw what looks like a restaurant. They do not know your walls, your drains or the rules where you trade, and nothing in the picture tells you so.
Used well, AI saves money in a different place. You walk into the first designer meeting knowing your seat target, what your menu needs from the kitchen and the three looks you like, so paid hours go on drawings instead of on discovering what you want.
Write the brief before you generate a single image
Image tools give you what you ask for, so the first job is knowing what to ask. A one-page brief also becomes the first thing you hand a designer. Ask ChatGPT or Claude to interview you rather than write from a blank page, because the questions you cannot answer turn out to be the most useful output.
Interview me, one question at a time, to write a one-page design brief for a restaurant.
Ask about: the concept in one sentence; service style (table service, counter, takeaway);
target seats and average spend; peak covers per hour; menu type and cooking methods
(gas, fryers, ovens, open flame); budget range and timeline; the unit's size, doors, windows
and existing services (gas, extraction, drainage); the neighbourhood and customers;
three venues I like and what I like about each; things I must avoid.
After the last question, write the brief in under 400 words and list every question I could not answer.Here is the sort of brief that comes out, for an illustrative 40-seat bistro:
Concept: a neighbourhood bistro serving seasonal plates and natural wine, relaxed but not casual. Service: table service, a short bar for walk-ins, takeaway pick-up by the door. Target: 40 seats plus six bar stools, average spend about $45 a head, peak of 28 covers an hour. Kitchen: gas hob, two ovens, one fryer, no open flame in view. Unit: ground floor, about 1,100 square feet (100 square metres), one front door, one rear door to a shared yard, two windows, a column near the middle. Budget and timeline: to be agreed after concept; open in six months. Likes: warm timber, low lighting, a visible bar. Dislikes: loud music, bare concrete. Unanswered: extraction route, whether the rear door can serve as a delivery entrance, how many toilets are required.
The "Unanswered" line is the valuable one. Each item on it is something a designer would otherwise bill you to discover, and each belongs on your pre-designer checklist further down.
Mood boards: three rounds, then a style sheet
Image tools are good at atmosphere: materials, colour, lighting, textures and furniture styles. Ask for those, and nothing else. Run three rounds.
Round one is exploration. Generate about a dozen images across three directions, such as warm oak and green tile, whitewashed walls and terracotta, or dark panelling and brass. Ask for no people and no readable text, because lettering in generated images is often garbled and people distract from the room.
Interior concept image of a 40-seat neighbourhood bistro, evening mood.
Warm oak tables, deep green tiled bar front, low brass pendant lights, linen-coloured walls,
banquette seating along one wall. Soft, dimmable lighting. No people, no text or signage, no logos.
Wide-angle view from the front door.Round two is scoring. Rate each image from one to five on five questions: does it fit the concept, could you build it on your budget, can staff clean it, is the seating comfortable for a ninety-minute dinner, and does the light flatter food and faces. An illustrative scoring grid:
| Image | Fits concept | Buildable on budget | Easy to clean | Comfortable seating | Light for food | Total |
|---|---|---|---|---|---|---|
| A: oak, green tile, brass | 5 | 3 | 4 | 4 | 5 | 21 |
| B: whitewash, terracotta | 4 | 2 | 3 | 5 | 4 | 18 |
| C: dark panelling, brass | 3 | 5 | 5 | 2 | 3 | 18 |
Image A wins, but its three on "buildable" is a flag, not a verdict. The tiled bar front is the expensive part, so the useful note for the designer is "keep the green, ask for a cheaper material". Scoring turns "I like it" into something a designer can act on.
Round three is the style sheet. Pick your top six images, upload them and ask the AI to describe what they share: materials, colours, lighting warmth, textures, furniture. Then edit the answer. A realistic first pass might say "reclaimed solid oak throughout", and your edit turns that into "oak look, solid or veneer, price to be tested", because a material the AI drew freely is a budget and supply question for a person.
Which tool? The ChatGPT pricing page lists limited image creation on Free and more on Go, Plus and Pro. The Gemini app's free image downloads are 1K with a visible sparkle watermark, and 2K on an AI plan. Canva's free plan allows up to 20 AI uses a month. Midjourney lets you build a Moodboard from your own reference photos and apply it to prompts; its help article says Moodboards work with version 6 and later and does not say which plans include them. Limits change often, so check each plan page. A free tier is enough for a first board; the comparison of concept image tools by cost goes into the paid ones.
Layout ideas: ask for zones and flows in words first
Do not ask an image generator for a floor plan. Ask a text assistant for ideas about where things go, in words, and keep the drawing for a person who can be held to the dimensions. Measure first: wall lengths, door positions and widths, window positions, ceiling height, columns, and where gas, water, drains and an extraction route could run. Sketch it on paper, photograph the sketch and put the fixed points in the prompt.
Here is a restaurant unit: about 11 m by 9 m. Front door in the middle of the short south wall,
rear door in the north-east corner, two windows on the south wall, one column roughly central,
drains along the north wall. Target: 40 dining seats plus a 6-stool bar. Menu: gas cooking,
one fryer, no open flame in view.
Propose three zoning options in words only, with no drawing: where the kitchen, bar, dining,
toilets, waiting area, deliveries and bins go. For each option list what is adjacent to what,
the path a customer takes from door to table, the path of a delivery from the rear door to storage,
and the path of dirty dishes from table to wash-up. Mark every assumption you made about dimensions.A typical answer reads sensibly and still hides problems. Here is an illustrative excerpt:
Option B: kitchen and wash-up along the north wall, with the rear door as the delivery entrance into dry storage. Bar opposite the front door so arriving guests see it first. Toilets in the north-west corner beside the kitchen. Dining in the middle and along the windows. Dirty dishes return along the east wall to the wash-up.
Now apply the fixes an owner can spot without a qualifying degree. The toilets sit beside the kitchen door, which a designer will want to move or separate properly. The dirty-dish route along the east wall passes the rear door where deliveries arrive, so two messy routes meet. A bar facing the front door means guests queuing for a drink cross the route to the tables. And the extraction route, the biggest unknown on your list, is not mentioned at all.
Choose the option with the fewest crossed paths. Then trace three routes with a pen on your own sketch: the customer, the delivery and the dirty plate. Wherever two lines cross, mark it. Those marks are the first thing worth showing the designer.
The kitchen sequence is not something to improvise. Commercial-kitchen guides from equipment suppliers agree on the basic order: receiving, storage, prep, cooking, pass, service, dish return and warewashing, with deliveries entering near storage and dish return kept away from the food pass. They also agree on a point worth underlining: a rear door that serves as an exit should never double as a storage area. Ask any layout, human or AI, to prove it respects both.
What a pretty render never shows
None of what follows is a reason to avoid the images. It is a reason to treat them as pictures. A reviewer can only check what is drawn, and the drawn part is the easy part of a restaurant.
| The picture cannot tell you | Why it matters | Who settles it |
|---|---|---|
| Exits and escape routes: how many, how wide, which way doors open, how far the furthest seat is from one | Fire and building rules decide, and a blocked or missing exit stops the whole project | Designer, with the fire and building officials who approve plans |
| Capacity: the legal maximum number of people in the space | It caps the number of people you may admit, whatever the picture shows | Designer and local officials |
| Step-free access, door widths, turning space and an accessible toilet | Expected or required in many places and costly to add after the build | Designer; an access specialist if the unit is awkward |
| Kitchen separation: raw and ready-to-eat food, handwash basins, clean and dirty routes | Food-safety inspectors are likely to look at these before you open | Kitchen designer and the food-safety authority |
| Extraction, gas and fire suppression over cooking equipment | Can be a large cost and a long approval, so ask about it first | Mechanical engineer, designer, fire officials |
| Drainage, grease management and hot water capacity | Wrong drains or an undersized water heater are expensive to correct | Plumber or engineer; your landlord for shared drains |
| Deliveries, bins, cold and dry storage, staff changing | The back of house decides whether service runs smoothly | You and the designer |
| Sound, draughts and comfort | Hard surfaces make a room loud; a door that opens onto a table is a daily irritation | Designer; acoustic advice for larger rooms |
| Service points: POS stations, runner routes, drinks and cutlery stations | These make a layout work during a rush, and images never include them | You, with the designer |
The seat-count test on one render
Suppose the owner has fallen for one render. It shows ten two-seat tables, eight four-seat tables and a bar with eight stools. Count: 20 plus 32 plus 8 is 60 seats. The brief asked for 46: 40 at tables plus a six-stool bar.
Now the room. The shell is 1,100 square feet. Suppose the designer's first sketch gives 400 to the kitchen and back of house, 100 to the entrance, waiting area and toilets, and leaves 600 for dining and the bar. Suppose, too, that the designer plans on 13 square feet for each seat, bar stools included. That figure is invented for this sum; the real one varies with service style and local rules, and your designer will supply it. Then 600 divided by 13 gives 46 seats, which matches the brief. Sixty seats in 600 square feet is 10 square feet a seat, almost a quarter less room than the allowance. The render has probably got there by drawing aisles nobody could walk down and tables a server could not reach across.
Three more checks take about five minutes each:
- The door-swing check. Every door in the image must open without hitting a table, a chair or a queue.
- The dirty-plate trace. Follow one used plate from a table to the wash-up with a pen. If it crosses the pass, the bar queue or the front door, the layout fails.
- The lettering rule. Any text in the image, from a menu board to a sign, is decoration. Do not read anything into it.
If you want an image tool to restyle a photograph of the empty unit instead of inventing a room, watch what it moves. Ask it to change finishes only, and check that walls, windows and ceiling height stay where they are. If the architecture shifts, you have lost the one thing you were trying to keep. AI room visualisation from client photos shows how designers handle this, and AI floor plan generation for architects explains where generated plans fail a real brief.
Three restaurant types, three different weak spots
The same checks land differently depending on what you are opening.
A 20-seat takeaway-led pizza counter. Seating images matter least. The things an AI render will not show are the queue, the pick-up shelf and where delivery riders wait without blocking the door. Trace the path of an order from till to oven to hand-over, and count how many people can stand in the shop before the door is blocked.
A 70-seat bar-led room. Sightlines and noise dominate. Check that the bar staff can see the door and the tables, that glass-washing has its own space away from food prep, and that the bar queue does not cross the route to the toilets. Hard surfaces that look wonderful in a render can make the room painfully loud.
A bakery café. The display counter, the ovens and the early deliveries decide everything. The render will show a tempting cabinet and cakes, but not the flour deliveries at 5 am, the trolley route or where the proving racks go. Put the bake-off equipment and its extraction needs at the top of the checklist.
Your pre-designer checklist, filled in for a 40-seat bistro
This is the part to keep. Fill it in before the first designer meeting. The answers below are for the illustrative bistro; blank items are questions for the designer.
Site facts
- Lease length and any break clause: 10 years, break at year five.
- Permitted use: confirm in writing that the unit may operate as a restaurant. Alterations need the landlord's consent: yes.
- Services present: mains water and drains yes; gas to be confirmed; extraction route to the roof or the rear: unknown, ask.
- Entrance: one small step at the front door. Ask about ramp and door options.
Concept and numbers
- One-sentence concept, seat target (40 plus six stools), average spend ($45), peak covers per hour (28), opening hours and expected share of takeaway.
Kitchen
- One-page menu, cooking methods, equipment wish list, days of stock held (three days of dry goods, two of fresh) and deliveries per week (four).
People
- Staff at peak: three in the kitchen, four on the floor, one on the bar. Areas needed: lockers, a changing space, an office corner, a staff drink and a rest area.
Customers
- Waiting area for walk-ins, space for pushchairs, a route for wheelchair users to the bar and a toilet, any outdoor seating.
Approvals the designer must list, with who submits each
- Building and fire approval, food-business registration, any alcohol licence, signage consent, outdoor-seating permission, and any extraction or planning permissions. Ask which apply to this unit and where you are.
Money and time
- Budget bands for design fees, building work, equipment and a contingency; target opening date; any fixed deadline, such as the end of a rent-free period.
What to send the designer, and what to ask for back
Send a short package: the brief, the mood board and style sheet marked as inspiration, your measured sketch with the fixed points and crossed paths, and the checklist with its questions. Say plainly that the layouts in the images are not buildable drawings. Here is an illustrative covering note:
Hi [designer], we're taking the unit at [address] for a 40-seat bistro and would like your help from concept stage. Attached: a one-page brief, a mood board made with AI image tools (inspiration only; the layouts in the images are not drawings and I know they don't work), my measured sketch with the fixed points, and a list of questions. Could you tell me your fee and scope for concept and developed design, which approvals the project needs and who submits each, the maximum occupancy you'd expect for the space, and the three biggest risks you can see from the sketch?
Ask for these five things back, in this order. A scaled plan showing seat count and circulation. The list of approvals and who signs each off. An equipment schedule showing the services each item needs (gas, power, water, drains, extraction). A budget with a contingency line. A timeline that includes approval lead times. If the designer can name the three biggest risks from your sketch in the first meeting, the preparation has paid for itself.
The same habit of briefing a professional with AI-made material works in other creative fields, and briefing a brand designer with an AI-made mood board covers it from the branding side. If a general fit-out plan is what you need next, ChatGPT for planning a shop fit-out from brief to Gantt chart gives the project view.
Rights, privacy and look-alike risks
A few cautions are worth a minute of your time:
- Do not name a real venue in the prompt. "In the style of" a named restaurant can produce a close copy of someone's look. Describe materials, light and mood instead.
- Build any Midjourney Moodboard from photos you took or have permission to use. A board of other people's interiors invites the same problem.
- Keep the lease out of free chat tools. Do not paste landlord correspondence or building plans into a consumer chat. Use a business plan, or paste only the facts you need.
- Remember what you own. OpenAI's and Anthropic's terms assign you whatever rights they have in outputs, "if any", so ownership of a generated look is not guaranteed. Using AI-generated images in marketing legally covers the wider question.
A weekend plan and what it costs
These are planning estimates, not measurements. An hour for the interview and the brief. Three hours for the three image rounds, including scoring. Two hours for the zoning options and the three-route trace. An hour for the checklist and the covering note. That is about seven hours and, on free tiers or one month of a $20 list-price plan, between nothing and $20 in tools. What it buys is a first meeting where the designer starts from your numbers. Ask them how they would scope the project given the brief; their answer tells you whether the AI-made material is helping.
Questions owners ask before showing AI images to a designer
Can AI check my restaurant layout against fire and building rules?
No. It can list questions to ask, but exits, occupancy, access and extraction are decided by a qualified designer and the local officials who approve plans. An AI plan that looks compliant is a drawing, not an approval, so treat every layout it gives you as a sketch to be checked.
Can I give AI images to a builder to price?
Use them to show style, not to price. Builders and suppliers price from scaled drawings and a specification, which the designer produces. Hand over the mood board as inspiration and say clearly that the layouts in the images are not buildable drawings.
Which AI image tool is best for restaurant interiors?
Run one prompt through two tools and judge which follows your materials and lighting more faithfully, not which first image looks prettiest. Free tiers of ChatGPT and Gemini are enough for a first mood board; Midjourney's Moodboards are worth a look if you have your own reference photos.
Should I tell the designer I used AI?
Yes, and label the images as inspiration. A designer needs to know which parts of the brief are your considered choices and which are just pictures you liked. Being open also lets you ask them to point out what the images get wrong.
Further reads
- Best AI for Interior Design Mockups: Client-Ready Images for a Studio — Client-ready mock-up images, and how a studio uses them.
- AI for Retail Store Design: Layout and Display Ideas — The same approach for a shop floor, with display ideas.
- How Much Does AI Cost a Small Restaurant Each Month? — What AI tools cost a small restaurant each month once it opens.
- AI Marketing for Restaurants: A Four-Week Plan Owners Can Follow — What to do about marketing while the fit-out is under way.
- AI Event Design for Planners: Mood Boards and Layouts Clients Approve — Use image tools for mood boards and rough layouts, run a clean client approval round, and keep every measurement human-checked.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Midjourney Moodboards help article, ChatGPT pricing page, Canva pricing page and commercial-kitchen layout guides from equipment suppliers (all checked 28 October 2026), Google's Gemini app help pages on image downloads, and OpenAI's and Anthropic's terms on output ownership.