1. Home
  2. Blog
  3. Real estate

3D Real Estate Websites: Selling Homes Before They're Built

A real-time building model can be the menu, the price list and the mood film at once. Here is how it works, what it needs, and how to keep it fast on phones.

ASFARI Lab5 min readOctober 2026

A brick industrial-heritage residential building at dusk with scattered windows lit warm coral

A 3D real estate website uses a live, real-time model of the building as the main way to explore and buy. Instead of a gallery of renders and a separate price table, the buyer rotates the building, hovers a window to see that exact home's rooms, size, price and status, and filters until only the matching homes light up. For off-plan projects it is the closest thing to standing outside the finished building at dusk.

It is also the most demanding option.

Why the building is the best sales tool

A buyer's first questions are spatial: which side faces the sun, what the courtyard looks onto. A flat list cannot answer those; a model answers them for every home at once.

Three interactions do most of the work:

  • Window as home. Each window or balcony is linked to one unit. Hover or tap shows rooms, area, outdoor space, price and status.
  • Light up available homes. One toggle dims everything sold or reserved and lights what is left. It doubles as a scarcity signal without any copywriting.
  • Day to dusk. The same building in morning light shows orientation and views; at dusk it becomes the emotional hero shot.

Our Tallinn example, as an illustration

To test these ideas, we built an example residential project in Tallinn: a brick industrial-heritage building rendered in real-time 3D at dusk. What it includes:

Feature What it does
Every window is a home Hover tooltip with rooms, area, outdoor space, price and status
"Light up available homes" Toggle that lights only unsold homes
Facade elevation picker Choose a side of the building; a filterable table below follows
Neighbourhood in walking minutes Everyday places shown as walking minutes, not adjectives
5-stage construction timeline From permit to handover, so buyers see where the project is
Sales contact form Rooms and price slider, with the selected home pre-filled
Low-end phone fallback A pre-rendered loop video replaces the live model

It is an example project, not a launched development, so there are no results to report. It shows how the pieces of an interactive apartment selector change when the selector is a live 3D scene instead of a static render.

What a 3D real estate visualization needs

Diagram showing data flowing from the price list and CRM into the apartment selector and lead form
The model is the interface; the data still comes from the sales team's price list or CRM.

The model

Start from the architect's CAD or BIM model, then rebuild it for the web. A construction model is far too heavy; you keep the facade, windows, balconies and a simplified site, and bake detail into textures. The deliverable is usually glTF, the standard web 3D format, compressed with Draco for geometry and KTX2/Basis for textures.

Each window or balcony is a named mesh whose name is the unit ID. That single convention is what makes window-as-home possible.

The data

The model never stores prices. At runtime, the site loads the unit table (from a CRM, a headless CMS or a shared sheet), matches unit IDs to meshes and colours them by status. When sales reserves a home, the building changes on the next load.

The light

Day-to-dusk can be faked with two baked lighting sets and a blend, or done with real-time lighting. Baked is lighter and more predictable on phones; real-time is more flexible for sun studies. We usually bake.

Keeping an interactive 3D building fast

3D has a cost, and phones pay it. Utsubo notes that Three.js core is around 600 KB minified before any scene code, and adding helpers, post-processing and a model can push you past 3 MB. Google's "good" LCP is 2.5 s or less (web.dev).

What we do:

  1. HTML first. Headline, prices, availability and the unit list render as normal HTML before the 3D boots. Utsubo's advice is to "ship the page as HTML, hydrate the canvas after."
  2. A poster frame. A still of the building at dusk is the first paint and the LCP element.
  3. Lazy-load the scene after first paint, and only when the hero is in view.
  4. Device check, then fallback. Low-end phones get a pre-rendered loop video and a 2D facade selector with the same data. Nobody gets a blank canvas.
  5. Budget the scene. Set limits for model size, texture memory and draw calls, and test on a mid-range Android phone, not a studio laptop.

Our full checklist lives in 3D website performance.

When a 3D real estate website is worth it

Situation Our view
Off-plan, 40+ homes, strong architecture Yes, the model is reused across web, sales gallery and film
Heritage or landmark building Yes, the building is the story
Under 15 homes, simple block Probably not; a good render selector is enough
No reliable price list or status process Fix that first; 3D on stale data is worse than no 3D

Whatever you pick, the 3D layer sits on top of a complete real estate development website: plans, location, progress and contact still need to work without it.

Questions people ask

Sources

LET’S OPEN A NEW TAB

What are you
thinking?

What’s on your mind?

This prepares an email to lab@asfari.studio. You review and send it from your email app.