Skip to content
Enrique Bay ESEN
Projects · 3D model · 2026

My home’s digital twin

My house in 3D, at true scale, to decide which furniture to buy and where to put it without guessing.

Time
October to November 2026, on and off. Hours: to be confirmed
Models
Claude, GPT Image and Gemini
Cost
To be confirmed
Status
Private. My family uses it to decide

What it is

We are furnishing a house and I did not want to buy blind. I built a true-scale 3D model you can walk through on a phone, where every candidate piece goes in at scale, with its price and its link.

It is my home, so I do not publish floor plans or walkthroughs. What I do share is the method, documented as a kit so it can be repeated in another space.

How it was made

  1. Capture

    The original floor plan, video of every space and drone shots. Plans drift from reality over time: every difference gets written down.

    Floor plan · video · drone

  2. Data before images

    Walls, openings, columns and levels live in a data file where every measurement carries its source and how reliable it is.

    Claude · Python

  3. Calibrated cameras

    To compare a render with a photo you have to stand in the same spot. Cameras are solved from points that match between the photo and the model.

    Python · OpenCV

  4. Real furniture

    Every piece goes in with the store’s measurements. When the store publishes a 3D model, that is the one I use.

    Blender · GLB models

  5. Whoever builds does not grade

    Each render is judged by a fresh agent that did not see the work: yes or no, and at most five problems with their fix in numbers. The grade is the worst view.

    Claude · Blender

  6. Decide from the phone

    Options arrive as side-by-side images with prices, ready for the family chat. For matters of taste, there is always an A and a B.

    three.js · spreadsheets

Models and tools

Claude
The data model, the scripts and the review of every render
Blender 5.3
Photoreal rendering, driven from Python
three.js
The walkable tour in a single page
OpenCV
Textures taken from real video
GPT Image and Gemini
Polishing mockups and cleaning photos
Playwright
Checking the tour works on a phone

The creative part

The rule that changed everything: measurements come from the floor plan and the scans, and looks come from the photos. Never from an AI model’s imagination.

The first version looked like a store catalog. My family said so bluntly, and they were right. Today the twin is more useful for ruling things out than for falling in love with a render.

Time and cost

Time
October to November 2026, on and off. Hours: to be confirmed
My hours
To be confirmed
Subscriptions
Claude and ChatGPT, which I already pay for. This project’s share: to be confirmed
Blender, three.js and Python
Open source, no cost

More projects

See all