For an interior, open ChatGPT's or Gemini's image editing, upload the client's room photo and list what must not change before what should. For an exterior, Hover Design fits better because it applies real manufacturer products to the house. None holds exact dimensions or finishes, so label every picture a concept and let the quote's specification stay in charge.
The tools are the easy part. What decides whether a rendering wins the job or costs you margin is the photo you start from, the details you refuse to let the AI change, and the wording under the picture. OpenAI's developer guide to its image model says the model "may have difficulty placing elements precisely in structured or layout-sensitive compositions", which describes a kitchen well. Hover's own Design page says final appearance "can still vary based on lighting, installation, and exact product selection". Both vendors are telling you what belongs in the caption.
Which route fits an interior, an exterior or a measured model
Four routes are worth knowing, and a fifth, full render software, is what most pages that rank for this keyword are selling. Prices and limits below were read on vendor pages on 29 October 2026. They move, so treat them as a starting point and check the page before you build a habit on one.
| Route | Best for | What you give it | Cost to try | Where it stops |
|---|---|---|---|---|
| ChatGPT image editing | Interiors: kitchens, bathrooms, living rooms; colour and style options | The client's photo, a written brief, optionally photos of products as references | Free plan to test; limits differ by plan, so read the plan page. Images upload at up to 20 MB each | Drifts on scale and small details; no product catalogue behind it |
| Gemini image editing | The same jobs, as a second opinion | Upload an image and ask for edits | Free; downloads are 1K without a Google AI plan and 2K with one | Google's help page says to use your discretion before you rely on or publish generated content |
| Hover Design | Exteriors: wall and trim materials, windows, doors, garage doors, shutters, paint | One photo for 2D designs; a scan for the 3D model | First three projects free, no card; Design itself is included from the Pro plan at $999 a year | Exterior examples only; no accuracy figure is stated |
| ServiceM8 AI image generation | Firms already running ServiceM8 | A site photo, to help clients "visualise the outcome you're quoting" | Ticked on all five plans, Free included, on its pricing page; no limits stated | Nothing says whether it counts against the Free plan's 10 AI uses a day; ask for a demo |
| Render software such as SketchUp | Agreed layouts where dimensions matter | A measured model | SketchUp Pro $33.25 a month billed annually; Studio, the only core plan listing V-Ray, $71.58 | Someone has to build the model, which is far more time than a quote usually justifies |
My rule of thumb: an interior photo and a quote due tomorrow means ChatGPT or Gemini; an exterior with real product choices means Hover Design; a layout the client has approved and you must now build to means a measured model. I have not seen a published test that ranks ChatGPT against Gemini on keeping a room's geometry intact, so run the same brief through both on one of your own old jobs and keep whichever respects the room better. If you run a design studio rather than a building firm, the studio version of this comparison is in Best AI for Interior Design Mockups.
Prepare the client photo before any tool sees it
Most bad renderings are bad photos. A wide, tilted, cluttered shot gives the model a room it has to guess at, and it guesses. These steps take ten minutes and are worth doing every time.
- Ask for three photos rather than one: from the doorway, from the opposite corner, and a close shot of the area that is changing. Landscape orientation, phone held level, daylight plus the room lights on.
- Ask the client to clear worktops and move bins and boxes. The model treats everything in the frame as part of the room and will happily keep a bread bin in your new design.
- Skip panorama and portrait modes. Panoramas bend straight lines and portrait mode blurs edges, and both make the geometry harder to preserve.
- Ask for the original files by email or a shared folder rather than screenshots. Some messaging apps shrink photos, and a small, soft image comes back as a small, soft rendering.
- If you visit, hold a tape or a one-metre rule against a wall in one frame. You will use it to check scale later.
- Write the real dimensions beside each photo: wall lengths, ceiling height, window and door sizes. This sheet is the source of truth, not the picture.
- Crop out people, framed photos, letters and the house number before anything is uploaded, and keep a line in your quote terms saying the client agrees to their photos being processed with AI tools.
File limits are rarely the problem. OpenAI has listed 20 MB per uploaded image for ChatGPT, and Google's help page allows 10 files per prompt in the Gemini app at up to 100 MB each, so phone photos sit well under both. Name each file with the job reference and the wall it shows, for example J1042-kitchen-long-wall.jpg, so the right photo goes with the right brief and the finished picture can be traced back to its source.
A galley kitchen from photo to concept image
Here is the whole routine on one job. A three-person remodelling firm gets an enquiry for a galley kitchen, 3.6 m by 3.0 m (about 12 ft by 10 ft). The window is on the long wall above the sink, the door on the short wall, and the client wants sage-green shaker units, a light stone-effect worktop and an oak-effect floor. The quote will land near $18,500 with roughly $4,000 gross margin. These figures are illustrative.
The first prompt puts the fixed things first and the changes second. Nothing about it is clever; the structure does the work.
You are editing a photograph of a real kitchen for a quote.
Keep the camera angle, room shape, ceiling height, window
position, door position and radiator exactly as in the photo.
Do not add or remove windows, doors or walls.
Change only these things:
1. Base and wall units: shaker-style doors, muted sage green,
brushed brass handles.
2. Worktop: light stone-effect surface.
3. Floor: mid-oak wood-effect vinyl.
Leave the appliances where they are. No text, logos or people.
Photographic style, natural daylight.
An illustrative first result looks convincing and has three faults. The cabinet run on the long wall appears longer than the real 3.6 m because the model has narrowed the window to fit more units in. It has added a wall unit beside the window that is not in your plan. And the sage reads closer to grey, because colour in a generated image follows the lighting in the photo, not the paint chart.
Correct the faults in a short follow-up rather than rewriting the whole brief. Asking for everything again tends to move the things you had already fixed. If a correction does not hold, send it on its own.
Same image. Restore the window to its original width.
Remove the extra wall unit beside the window. Keep everything
else exactly as it is.
Two rounds like this is normal. Then run the drift check in the section below, attach the picture with its label, and move on. The time budget for the whole routine, again illustrative: ten minutes of photo preparation, fifteen minutes of prompting, ten minutes of checking against the dimension sheet and five minutes to label and attach. Forty minutes at an internal cost of about $35 an hour is roughly $23 of time, and the tool itself can be free. Against a job with $4,000 of margin, one extra win from thirty quotes would cover years of a $20 monthly plan. I would not promise that to anyone, because a picture only helps when the quote behind it is sound, but the sum shows how little downside there is.
The same photo set can also feed the price. A picture says nothing about what is behind the plaster, so keep the scope lines in the quote itself; Kitchen Remodel Estimate Template: Line Items Written With ChatGPT covers that half of the job, and Remodelling AI Estimator: Test It Against Your Own Past Jobs shows how to test an estimating tool before you let it near a price.
Exteriors: where Hover Design earns its fee
Going by every example on its page, Hover Design is built for the outside of a house. You upload one photo and try materials on the 2D image; after a scan, you can view the same design on a 3D model. Hover says the products come from "30+ brands" and "tens of thousands of products", and you can apply materials to individual facets, split walls and trim, and swap doors, garages and windows. That is the difference from a general image tool: the cladding in the picture is a product you can name and order.
The pricing is pay-per-project on the Starter plan, with the first three projects free and no card needed. After that, an average exterior scan costs $99 on Starter, a simple one $59 and a complex one $139. The Pro plan is $999 a year, takes $20 off every project and is the plan that includes Design; Starter lets you "try Design and estimate with real brands" but does not include it. Business is $2,999 a year with a flat rate per project. A quick sum: at $20 off per project, Pro repays its fee through discounts alone at about 50 scans a year ($999 divided by $20 is 49.95). Below that, the case for Pro is the Design tool, not the discount.
Take a four-person cladding and windows firm that quotes 12 exteriors a month. On Starter, the first three scans are free, then nine at an average $99 comes to $891 in the first month. On Pro the same nine cost $79 each, $711, plus the $999 yearly fee. Over a year of 144 scans, 141 of them paid, Pro's discount is worth $2,820 against its $999 fee, so at this volume Pro wins on price before Design is counted: about $12,100 a year in total against about $14,000 on Starter. The bigger number is the scan spend itself, which belongs in the quotes it supports. Hover lists Measurements and Estimates as separate items built on the same 3D model, and its pricing table shows Estimates starting at Pro, so ask what your plan includes of each before you decide.
Two limits matter. Hover's page gives no accuracy figure for the designs, and it says the final look varies with lighting, installation and exact product selection. And Hover's separate Interiors product is a measurement tool: a phone scan that produces measured floor plans, photos and a 360-degree walkthrough. Its page describes measuring and walking through a room, not drawing a finished one. For interiors you are back to a general image tool.
What no image tool keeps: scale, structure and finishes
A generated picture is a plausible room, not a surveyed one. It may keep your window and move the radiator, shift a doorway by a hand's width, or make a worktop run longer than the wall. Run this five-point drift check against your dimension sheet before any image leaves the building.
| Check | How to do it | Pass mark |
|---|---|---|
| Openings | Count windows and doors in the original and the render; compare which wall each sits on | Same number, same wall, similar width |
| Ceiling and floor lines | Compare the top of the door frame to the ceiling line in both pictures | The gap has not grown or shrunk |
| Scale of the main run | Print both; measure the window width and the cabinet run; compare the ratio to your sheet | Within about 10% (my own rule of thumb, not a standard) |
| Fixed services | Look for the radiator, flue, sink waste and consumer unit | Where they are on site, or removed only if you are moving them |
| Colour and sheen | View on a phone and on a print next to a physical sample | The nearest sample is the one you will specify |
On the galley kitchen, the first render failed the third check: window about 30% narrower than the original, so the units ran longer than the wall allows. The second render passed all five. Keep a one-line note of the result in the job file, because it becomes your evidence if the client later says the picture showed something else.
Finishes are the second limit. Gloss levels, grain direction, grout lines, and the exact shade of a painted door do not survive generation. The honest way to handle this is to name the real product in the specification schedule, by manufacturer reference, and to let the client hold a physical sample. The picture shows mood and layout; the schedule shows what you will actually fit.
Pixel size is the practical limit that people forget. A half-page picture in a quote prints about 7.5 inches wide. OpenAI recommends 1536 by 1024 pixels for a landscape image from its image model, which prints at about 205 dots per inch at that width (1,536 divided by 7.5 is 204.8). Gemini's 1K downloads usually do for an email, and its 2K downloads, available with a Google AI plan, are the safer choice for a printed page. Decide the size before you build your quote template.
Wording that keeps the picture honest
Put the label under the image, not in a footer, and make it plain enough that a client skimming the page cannot miss it. Here are three versions to copy and adapt: the caption, the line in the quote and the line in the covering email.
Caption: AI-assisted concept image for discussion. Not to
scale. Layout and finishes are as set out in the
specification schedule, page 4.
Quote line: The images in this quote are concept images made
from your photographs with AI tools. They show the idea, not
exact dimensions, colours or products. The agreed scope and
specification schedule control what is supplied and fitted.
Email line: I have attached two concept images based on your
photos. Please treat them as a guide to style only; the
measured drawings and finishes come later.
Three habits support the wording. Name the real finishes in the schedule before the picture goes out, not after the client has fallen in love with a colour that does not exist. Keep the original photo and the final prompt in the job folder. And do not call the image a "visualisation of your new kitchen" or a "preview"; both words promise more than a concept image can.
Whether a picture becomes part of the contract depends on how your contract documents are written, and I am not the person to advise on that. If your terms list the quote and its attachments as the agreement, the image is inside it, so ask a lawyer to check your clause. How to Mention AI Use in Client Contracts and Proposals covers the wording from the contract side.
The island that did not fit: a mistake worth rehearsing
Here is how this goes wrong, in an illustrative case. A firm sends a render of an open-plan kitchen with a long island and four stools. The client loves it. At the survey, the room turns out to leave about 900 mm of walkway beside the island once the units are in, and the island has to shrink by a third. The client says the picture showed a different room. The firm eats a day of redesign and a margin conversation it did not need.
The fix is dull and cheap. The photo would have failed the drift check on scale if someone had put the tape measure next to the render. The quote would have carried a caption and a specification line reading "island: 1.8 m by 0.9 m, subject to survey". And the dimension sheet would have gone to the client before the picture did. Nothing here needs better AI. It needs the routine.
Score a tool on one finished job before you trust it
Pick a kitchen or bathroom you finished last year. Feed the tool the "before" photo and your final specification, and compare the result to your "after" photos. Score each tool from 0 to 2 on five criteria, then add them up.
| Criterion | 0 points | 1 point | 2 points |
|---|---|---|---|
| Kept the openings | Moved or invented one | Right count, wrong width | Matches the original |
| Kept the scale | More than 20% out | 10 to 20% out | Within 10% |
| Matched the finishes | Wrong materials | Right materials, wrong shade | Close enough to discuss |
| Followed the brief | Changed things you did not ask for | One stray change | Only the requested changes |
| Fixed in one follow-up | Needed three or more rounds | Needed two | Needed one |
Here is a filled-in grid, with invented scores to show the arithmetic. Tool A scored 2, 1, 1, 1, 1 for a total of 6 out of 10 on a bathroom job, because it kept the window and door but stretched the vanity run and got the tile shade wrong. Tool B scored 2, 2, 1, 0, 1 for a total of 6, because it kept the geometry better but added a plant and a mirror nobody asked for. Equal scores, different weaknesses, which is exactly why you test on your own work. Use 7 out of 10 as the bar for putting a tool into live quotes, and re-run the test every few months, since the models behind these tools change without notice.
When to hand the job to proper render software
A concept image is the right tool until the client says yes. Then the question becomes whether the layout is correct, which needs a measured drawing. Move to a model when the client approves a layout and the next step is ordering, when anyone who approves the work needs scaled drawings, or when you keep running the drift check and failing it on the same room. SketchUp Pro lists at $33.25 a month billed annually, with the V-Ray renderer only in the $71.58 Studio plan, and someone has to build the model.
That is why the pages that rank for this phrase mostly describe dedicated render products: they are describing the tool for the stage after the quote. For small firms, the sensible split is a quick concept image at quote stage, then either a measured drawing from your own team or an outsourced visualiser once the job is won. AI Garden Design Visuals in Quotes: Do They Help Close Jobs? weighs the same trade-off for landscaping, and the further reads below cover where outsourcing pays.
Questions remodellers ask about AI concept images
Do I have to tell the client an image is AI-made?
Yes, as plain dealing. Put 'AI-assisted concept image' in the caption and on the quote. Some tools add their own mark as well: free Gemini downloads may carry a visible one, so check what yours look like before they go into a quote. Some places also have rules on labelling generated images, so ask an adviser if you advertise with them.
Is it safe to upload a client's photo to a free AI tool?
Only after cropping out faces, documents and the house number, and after the client has agreed to it in your quote terms. Consumer ChatGPT and Claude plans let you switch off use of chats for model training, and business plans do not train on your content by default. Check the privacy setting in the account you use before uploading.
Does Hover Design work for kitchens and bathrooms?
Nothing on Hover's Design page says so. Every example it gives is an exterior: wall and trim materials, doors, garage doors, windows, shutters and paint. Hover's separate Interiors product measures rooms from a phone scan and produces floor plans and a walkthrough, not finished-room pictures. Ask Hover directly before paying for interior visuals.
How many versions should I put in a quote?
Two, at most three. Each extra image is another thing the client can hold you to and another round of checking for you. Show the option you recommend and one alternative, label both as concepts, and keep the full list of finishes in the specification schedule rather than in the picture.
Further reads
- AI Room Visualisation From Client Photos: A Designer's Workflow — The interior designer's version of this photo-to-image routine.
- Bathroom Remodel Estimate Template: Catch Missing Lines With AI — Catch missing lines before the picture raises expectations.
- Bathroom Fitter Marketing: Before-and-After Posts From Job Photos With AI — Reuse finished-job photos as before-and-after posts.
- AI Construction Design Software: What Renovation Contractors Can Draw — What drawing software adds once the layout is agreed.
- AI Renders vs Outsourced Visualisation: What Practices Save — When paying a visualiser beats doing it in-house.
- How Painters and Decorators Can Quote From Photos With AI — Quoting from client photos without a site visit.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Hover pricing, Design and Interiors pages; OpenAI's developer guide to image generation; Google's Gemini Apps help pages on creating and editing images and on uploading files; SketchUp plans and pricing page; ServiceM8 pricing page and feature overview. Prices and limits were read on 29 October 2026.