Simam AI Lab Our applied AI research division is now open. Visit the lab
Commissioned demonstration · client under NDA

Selling homes from a 3D of the building itself.

A client commissioned a marketing demonstration: could a residential tower be rebuilt in 3D from ordinary drone footage, and every home in it opened up on its own floor, on its own side of the building, with its view and its price? Seven clips in. A browser you can orbit, filter and export from, out. It runs on a phone with nothing to install.

The problem

A floor plan asks the buyer to do the imagining.

Homes in a tower are sold from plans, a price list and a render of the outside. None of those answers the question the buyer is actually asking, which is what this particular home, on this floor, on this side, will be like to stand in and look out of. A plan cannot show a view. A render of the lobby cannot show the ninth floor facing the mountains.

Need 01

“Which one is mine?”

A unit number on a price list means nothing until it is the room you are standing in, at the height you will actually be at.

Need 02

“What will I see?”

The view is most of what separates two identical apartments, and it is the one thing a floor plan cannot carry.

Need 03

“What is still available?”

Availability lives in a spreadsheet that the thing the buyer is looking at knows nothing about.

What was built

Seven clips in, 347 homes out.

The tower and the blocks around it were reconstructed from drone footage flown near and far, then measured from that reconstruction and opened up as sample residences. Each one sits on its real floor and its real side of the building, carries its size, whether it faces the mountains or the city, its status and its price, and the whole register can be filtered and exported.

7drone clips, near and far, solved into one 3D
347sample homes placed on their own floor and side
26floors of the tower, opened up individually
220 mof surrounding ground reconstructed with it
The reconstructed tower in 3D with sample homes shown floor by floor
26 floors of sample homes, each one on its own floor.
A plan for the surrounding district drawn onto the real reconstructed streets
A sample plan for the district, laid on the real streets.
How it was checked

Measured from the model, then tested against frames the model never saw.

A reconstruction will happily agree with itself. The only way to find out whether it agrees with the world is to hold it against something it was not built from.

Validation

Every eighth frame, held back

The reconstruction was checked against frames that were deliberately not used to build it, which is the only check that can catch a model that has confidently invented geometry.

Scale

Anchored to a published height

The 3D is scaled from the building’s own published height rather than from anything the footage claims about itself.

AI answers

Kept only where the frames agree

Where a model was asked to read something off the scene, the answer was kept only where most frames agreed, and is marked as something for an engineer to confirm rather than a finding.

What the measurements are worth, stated plainly. Heights are scaled from the tower’s published height, and sources for that differ, so everything derived from it carries about fifteen per cent either way. That is normal for this kind of capture and it is enough to lay out homes, compare floors and judge a view. It is not a survey, it settles no boundary, and where a surveyed figure exists the surveyed figure wins.

What this costs to make

The reconstruction is the cheap part.

This matters more than it sounds, because it is the difference between a few hundred pounds of work and a commissioned experience. Turning footage into a 3D is mostly unattended machine time. What takes a person is deciding what the thing should do.

9 min 37 sa 29 second clip, start to finished 3D
52 mina 20 second walk-through, unattended
1.5 to 3 hof human attention on a clean job
about 14 MBdelivered, for a million points

Those are our own measured figures on our own hardware, across roughly sixteen hand-picked clips. They are a floor rather than a forecast: footage that arrives cold behaves worse. Of those sixteen, three failed outright and four needed an expert to rescue them, which is why we ask to see footage, or to advise before it is filmed, rather than quoting blind.

What it does not claim

The honest boundary.

The footage

Stock, not the client’s site

This was commissioned as a marketing demonstration rather than a survey, so it was built from stock drone footage of a real building. It is not a capture of the commissioning client’s own asset and does not represent one.

The content

Every home and price is invented

The building is real and the reconstruction is real. The homes, their availability, their owners, the prices and the surrounding plan are all samples, and the pictures of the future are illustrations rather than approved drawings. The building itself is not named.

The client who commissioned it is under non-disclosure and is not named, and nothing on this page is their property: not the building, not the plan, not a price. What was delivered to them was the demonstration itself, which is the thing being shown here.

Your building, your site

Have something nobody can see whole?

The valuable part was never the renderer. It was the route from a drone flying for a few minutes to a buyer on a phone opening the exact home they are being sold. Tell us what you need people to understand and we will tell you honestly whether footage can carry it.

Start a project brief How it was built Back to case studies