From geodata to a realistic landscape

For a test Telenor commissioned from Merkle in September 2025, I used OpenStreetMap data, Three.js and Nano Banana to turn geodata into a realistic landscape.

September 1, 2025

Generated landscape with a fjord, forest, and mountains from the Telenor geodata test.
The result of the September 2025 test Telenor commissioned from Merkle.

In September 2025, Telenor hired us at Merkle to test how geodata could be turned into a realistic landscape. I built the workflow, from the terrain model to the final result.

From map to landscape

The test shown here uses OpenStreetMap data. I built and textured the terrain in Three.js, then used Nano Banana to make it look more like a real place. We also tested variants using the Google Maps API (application programming interface).

The test in action

The video shows the stages: a map, the basic 3D model, the textured terrain, and the realistic result.

Silent demo of the workflow I built at Merkle for a test commissioned by Telenor in September 2025, from map and 3D terrain to a realistic landscape. The on-screen labels identify OpenStreetMap, Kartverket, Geodata, and ComfyUI.

A test like this could be used to explore how a landscape might look from a chosen viewpoint, with the terrain model as the starting point.

Keep reading

GenAI / MCP / Productivity

How I connected my assistant to ComfyUI

I built Comfy2MCP so my coding assistant can run approved ComfyUI workflows, change configured prompts, and save the results locally.

Terminal demo showing local CrewAI agents requesting an Entur journey from Grorud to Røverkollen.
GenAI / MCP / Productivity

How I connected local AI agents to Entur

I built an MCP prototype using Entur's open APIs, CrewAI, and Llama 3.1 8B through Ollama to plan Norwegian public transport journeys.