AI Renders vs Outsourced Visualisation: What Practices Save

Yes, for early design options: AI render tools cost roughly $10 to $60 per seat per month, while visualisation studios' own 2026 price guides put a single outsourced still at about $250 to $5,000. The saving only appears where you were paying a studio for concept images. For planning, competition and marketing finals, outsourcing still usually wins.

The reason is that the two are different products. A studio builds a controlled scene from your model: the window count, roof pitch and materials are what you drew. Most AI tools interpret a view and return something plausible, and that something can quietly change the design. So the useful question is which of your images need to be accurate and which only need to be persuasive.

Follow me on Instagram@sagnikteaches

What a visualisation studio charges per image in 2026

Studios publish their own price guides, and they cluster in similar bands. Treat these as ranges to test your quotes against, not a market survey.

Connect on LinkedInSagnik Bhattacharya

Several of those guides say the same thing about what moves the bill: revisions, not extra camera angles. A still quoted at $900 can end up nearer $1,300 once the client changes the cladding twice. Keep that in mind, because it is also where AI helps even when you keep the studio.

Subscribe on YouTube@codingliquids

What the AI render tools cost per seat

Prices below are USD list prices checked in September 2026. Credits reset monthly on all the credit-based tools and do not roll over.

The last column matters more than the price column. Model-aware tools start from your geometry, so a render is recognisably your scheme. Midjourney produces atmosphere and mood very well, but it does not know your building, and every image drifts. For a fuller feature comparison, see the best AI rendering tools for small practices.

A costed year for a four-person practice

To see where the money actually moves, take an illustrative four-person architect practice doing residential extensions and small new-builds. Last year it bought 24 outsourced stills at an average of $900: 16 for early client presentations and 8 finals for planning submissions and the client's sale particulars. On top of that it paid for 10 extra revision rounds at $150.

That is a saving of about $13,000, and nearly all of it comes from one line: concept images the practice stopped buying. Now run the same sums for a practice that never outsourced concept work and just showed clients SketchUp massing. It has no $14,400 to remove. For that practice, AI renders are a new cost of about $2,500 in year one, and the return has to come from somewhere else: winning more of the jobs it presents, or getting client sign-off a meeting sooner. Both are real, but they are not savings on an invoice, so be honest about which practice you are before you promise the partners a number.

Once you know the visualisation line, the rest of the budget is simpler. What AI costs a small architecture practice each month covers the other tools that sit alongside it.

Costs that never appear on the tool's pricing page

• Curation time. You rarely get the image you want on the first run. Expect to generate 10 to 20 variations to find two worth showing, which is why the example above allows 1.5 hours per usable image.
• Checking against the drawings. Someone has to compare each AI image with the model: glazing bars, roof lines, the number of storeys, the neighbour's boundary. On a busy week this is the step that gets skipped.
• Credit overspend. AI-enhanced scenes, video and upscaling use more credits than a single image. Top-up packs exist on the Chaos tools, and a team that likes video can burn through a monthly allowance in days.
• Hardware for real-time renderers. D5 and Enscape run on your own graphics card. If the office runs on thin laptops, the true cost includes a workstation.
• The wrong tier for your size. Midjourney's revenue rule means most established practices cannot legally stay on the $10 plan, and the cheaper tiers make your images visible to other users. For a practice over the threshold, the real figure is Pro at $48 a month billed annually, or $576 a year, not the $96 a year the Basic plan suggests.

What a first AI render gets wrong, and the fix

The 1.5 hours per usable image is mostly spent correcting runs like this one. It is an illustrative first attempt on a two-storey rear extension, sent from the SketchUp view through a model-aware tool:

Prompt: Rear garden view, late afternoon in early autumn, soft low sun.
Larch cladding weathered to silver grey. Large sliding glazed doors,
dark bronze frames. Wildflower planting, mown path. Photorealistic,
eye-level, no people.

What came back (illustrative): a convincing image with the right cladding and good light. But the flat roof had become a sedum green roof, the three sliding panels had become four, a glass balustrade had appeared on the roof edge, and the neighbour's fence had grown into a brick wall. The client would have loved it, and asked for all four changes by name.

The fix took three more runs. Name what must stay ("flat single-ply membrane roof, no balustrade, three glazed panels"), turn up the tool's setting that controls how closely it follows the model geometry, and crop the reference view so the boundary sits outside the frame. Then compare the survivor with the model view before it goes anywhere near a slide.

Where the studio still earns its fee

Keep outsourcing, or render properly from your own model, when any of these apply:

• Someone will rely on the image to make a decision. Planning officers, a neighbour's objection, a developer's investment committee. An AI image that adds a dormer you didn't draw is worse than no image.
• The image will sell something. Brochure hero shots and sales particulars carry the practice's name and the developer's money. Studios control light, lens and materials to a standard AI tools do not yet match reliably.
• You need consistency across a set. Twelve views of one scheme, all in the same light, with the same brick. AI tools struggle to hold a scheme steady across images.
• You need motion. AI video is improving, but a studio walkthrough follows a camera path through your actual building.
• Nobody in the practice has time to learn a tool. A studio is also buying back hours. If your only designer is billing 35 hours a week, an hour and a half per image may cost more than the studio.

That last point is worth a quick sum, with illustrative figures. If the designer's time is billed to clients at $85 an hour, a usable exterior concept at 1.5 hours costs about $128 of fee-earning time, against roughly $800 from a mid-range studio, so AI wins easily. A joinery-heavy interior is a different animal: a kitchen with bespoke units, a feature staircase and the client's existing dining table can take six hours of runs and corrections, which is $510 of lost fees before the subscription. Against a $600 studio interior with two rounds included, that is close to a wash, and the studio's version will match the drawings. The pattern most practices end up with is AI for exteriors and simple interiors at concept stage, and the studio or a proper model render for anything with detailed joinery.

The consistency problem also shows up in a predictable way. A developer asks for six views of a terrace of four houses for a sales brochure. Run through an AI tool, the brick drifts from buff to red between views two and five, and the bay windows change width. Rendering all six from the model in D5 or Enscape keeps them consistent, because the materials live in the model rather than in a prompt.

How to brief the hybrid so revision rounds fall

The biggest studio saving comes from agreeing the look before the studio starts. Show the client two or three AI options and get a written choice. The covering email decides whether that choice is useful. A version that invites trouble: "Here are some images of how your extension could look, let us know what you think!" The client replies "love B, especially the roof terrace", and the roof terrace was never in the scheme. A version that works:

Attached are three mood options for the rear extension (A, B, C).
They show light, planting and cladding colour only. The roof,
glazing and layout follow the drawings, not these images, so
please ignore any differences in those.
Could you reply by Friday with the option you prefer, and anything
about the mood you would change? We will then brief the final
render for the planning submission.

Then send the studio a brief. Here is one filled in for the extension above (illustrative):

Project: rear extension and loft, two-storey semi-detached house,
  planning stage
Model: attached SketchUp file; camera positions saved as
  "Garden-01" and "Kitchen-02"
Agreed reference: attached AI image option B. The client chose this
  for mood, light and planting ONLY. Geometry must follow the model,
  not the reference.
Known differences from the reference:
  - reference shows four glazed panels; the design has three
  - reference shows a green roof; the design has a flat membrane roof
  - reference fence is brick; the actual boundary is a timber fence
Materials: larch cladding (weathered), dark bronze aluminium frames,
  grey membrane roof; product sheets attached
Time of day / season: late afternoon, early autumn
Deliverables: 2 stills at 4000 px wide, JPG and TIFF
Revision rounds included: 2; design changes after sign-off billed
  separately
Deadline: 14 working days, for the planning submission

The known-differences list does most of the work. It stops the studio copying a detail the AI invented and stops the client assuming the final will match the reference exactly.

Checks before a client sees an AI render

1. Compare it side by side with the same view from the model. Count windows, doors, storeys and roof planes.
2. Label it on the slide or sheet as an indicative image, not a representation of the final design.
3. Confirm the tool's terms let you use the output commercially at your plan level, and that your plan keeps the image private if the scheme is confidential.
4. Check that nothing identifiable crept in: a real brand on a shopfront, a recognisable neighbouring house, a person's face.
5. Save the prompt, the source view and the tool name in the project folder, so you can explain later how the image was made.

Here is how skipping step 2 tends to show up. A practice sends a client an unlabelled AI image of a kitchen extension with a vaulted ceiling and exposed oak rafters, because the tool liked them. The drawings show a flat plasterboard ceiling. Four months later, on site, the client asks where the rafters are, and the builder prices them as a variation. Nobody promised the rafters, but the image did. A one-line label, "indicative image; see drawings for the design", would have saved the conversation.

Client drawings and site photos are also client data. Before you upload them to a cloud tool, read the copyright and client data risks of AI in architecture, and if you are still deciding whether AI helps your practice beyond visuals, where AI saves time in a small architecture practice goes through the other jobs.

Pricing your own switch in an afternoon

Pull last year's visualisation invoices and sort every image into three piles: concept and option images, finals for planning or sale, and animations. Add the revision charges separately. A chat assistant does the sorting quickly if you paste in the invoice lines with client names removed:

Below are invoice lines from our visualisation studio for last year.
Sort each line into one of: CONCEPT (early options for client
meetings), FINAL (planning, competition or sales images), ANIMATION,
REVISION (extra rounds or changes). Give a table with the line, the
category, the amount, and a one-line reason. Then total each
category. If a line is ambiguous, mark it UNSURE instead of guessing.

[paste lines]

A typical reply looks like this (illustrative):

Check every UNSURE line yourself, and spot-check a few confident ones: assistants sometimes file a "stage 3 option" as a final because the word "planning" appears elsewhere on the line. Then price the AI route only for the concept pile, using your real staff cost per hour and 1.5 hours per usable image as a starting estimate. Take a 14-day trial on one model-aware tool and time yourself on two live projects before you commit to a year.

Keep a log during the trial, because memory flatters. Here is one filled in after two projects (illustrative):

That is three usable images from 395 minutes, about 2.2 hours each rather than the 1.5 assumed. Re-run the costed year at 2.2 hours and staff time on 16 concept images rises from $1,200 to $1,760, so the saving falls to about $12,400: still worth doing. The failed kitchen is useful too, because it tells you which kind of image stays with the studio. If the concept pile is small, the honest answer may be to keep the studio and spend nothing on AI renders at all.

Questions practices ask before cutting their visualisation budget

Can I submit an AI render with a planning application?

Treat AI renders as illustrative unless you can show the geometry, materials and viewpoint match the drawings. Model-aware tools help, but they can still add or remove details. For anything a decision-maker will rely on, use a render produced directly from your model, check it against the drawings, and keep a note of how it was made in the project file.

Do I need a new computer to run AI renders?

Not for cloud tools such as Veras or Midjourney, which do the heavy work on the vendor's servers. Real-time renderers such as D5 Render and Enscape run on your own machine and want a capable graphics card, so check the vendor's system requirements before committing, especially on older laptops.

Will a visualisation studio charge less if I send an AI concept image?

Often not on the headline price, but a clear reference image agreed with the client usually cuts revision rounds, which is where studio bills grow. Ask the studio how they price extra rounds and whether a signed-off reference changes their estimate, then compare two quotes on the same brief.

Further reads

• How Architects Use AI for Fee Proposals, Briefs and Paperwork — Use the time saved on visuals to tighten fee proposals and client briefs.
• How to Check AI-Generated Images for Errors Before Posting — A checking routine for AI images before anyone outside the office sees them.
• Can You Legally Use AI-Generated Images in Your Marketing? — What to know before AI images appear in your practice's marketing.
• AI Room Visualisation From Client Photos: A Designer's Workflow — How designers turn client photos into interior options quickly.
• Is Midjourney Worth It for Small Business Marketing Images? — A closer look at whether Midjourney earns its subscription.
• How to Calculate AI ROI for Your Business (Worked Example) — The general method for testing any AI spend against the time it saves.
• 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 Garden Design Visuals in Quotes: Do They Help Close Jobs? — When an AI-edited photo of the client's own garden helps a landscaping quote get signed, when it backfires, and a fair way to test it.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Chaos pricing pages for Veras and Enscape; D5 Render pricing and plan pages; SketchUp AI subscription help page and Trimble plans page; Midjourney plan comparison and Terms of Service; published 2026 rendering price guides from several visualisation studios. Checked September 2026.