Mapbox 3D buildings
Mapbox had no 3D landmarks. The job was a production system: recognizable, easy to customize, cheap enough to cover the world.
Decision
Maximum quality at a cost that can cover the world. Unique enough to recognize, neutral enough for any brand.
Mapbox Standard: lighting, landmarks, basemap. Open on Vimeo if the player does not load.
Problem
In 2021 Mapbox had no 3D landmark layer. Coverage had to be worldwide, and the target we set was to pass HERE. Contractors and freelancers, at launch, asked $30 an hour. A model took 6–8 hours. At that rate, quality and coverage could not both exist.
The model also had to satisfy three things that pull against each other, inside Mapbox's own rules for tiles, lighting, and how a building reads at navigation zoom:
- Recognizable. Unique enough that the building is the building.
- Brand-flexible. Easy to customize, and calm enough to sit under any customer's branding.
- Cheap to produce. A cost that still works when the catalog is the whole world, not a hero city.
What I built
A production system, before the team got big.
- One quality bar. What "done" means for silhouette, detail, mesh budget, and handoff, so a new person ships the same model the last person did.
- An in-house team. Trained people inside the company, most of them without a 3D background. Several still produce years later.
- Tooling around the work. Add-ons and checks that took repeated decisions out of the artist's hands. The hours dropped because the process got smaller, not because people worked faster.
Outcome
After the team scaled and the process, tools, and add-ons were unified, a model takes about 3 hours. That number is still going down. By 2026 the average cost is about $20 per model, at the quality and detail the product needs. The map carries about 20,000 unique 3D models.
The landmark layer in Mapbox Standard is that system running, not a set of hero buildings.