How Tjuvholmen in 3D was made

The outlines of the buildings come from OpenStreetMap, and the heights and terrain from the Norwegian Mapping Authority’s laser scan of Oslo. The quays, squares and shoreline follow the laser data, with the sea at mean sea level.

Aerial photos and pictures taken at Tjuvholmen were used as references for façades, materials and details. The model is approximate: it aims to give an impression of the district, not to be an exact copy.

Almost everything in the model is made by around 90 Python scripts that run in Blender: façades and windows, balconies and railings, the timber cladding on the quays, trees, boats and lights. The scripts can be run again, so a change in one place carries through everywhere.

Where something does not match reality, it is adjusted by hand in Blender, like the columns and the passage by the museum.

A photo taken from the bridge over the canal, towards the Royal Yacht at Filipstad, and the same view in the model. Drag the slider to see the difference.

The model is shown with three.js and WebGL, right in your browser, with nothing to install. It has around 600,000 triangles and is compressed, so it loads in a few seconds.

The light follows the time of day, with sunset, lit windows and the lights of the Oslo Tree in the evening. The water reflects the buildings, and the sea can be raised to storm surge level and to the year 2100.

Tjuvholmen in 3D was made by Vidar Vestnes together with Claude from Anthropic, using Claude Code. Claude wrote the scripts, controlled Blender and looked at the results, while Vidar decided what to make and checked it against reality.

The textures are originals made with AI. Artworks and murals are never copied from photos, but made in a similar style.