Rebuilding Central Warsaw in Voxels: The Birth of an Open-Source 3D City
Open-source project rebuilds central Warsaw in voxels, creating an interactive city sandbox from open data.
"Warsaw in voxels" is an open-source project that gained attention on Hacker News, rebuilding central Warsaw as an interactive 3D map in voxel style. Users can change the date, time, and weather, roam the city, and even click on pedestrians to control them, with interaction inspired by *Dungeon Keeper*. The project is built entirely on public data: OpenStreetMap provides the vector skeleton of buildings and streets, LiDAR data from Poland's GUGiK/Geoportal estimates building and tree heights, aerial imagery extracts colors, and Warsaw's urban tree database ties seasonal changes to real tree species. Released on GitHub, the author invites reuse, and this "open data to voxel city" pipeline can be transferred to other cities with similar data. Voxelization trades realism for lightness and interactivity, lowering the barrier to 3D urban visualization, though initial loading is slow, mobile experience is limited, and architectural detail is insufficient at voxel scale.
When a City Becomes Voxels: An Interactive Warsaw
On Hacker News, a project called "Warsaw in voxels" caught people's attention. It rebuilds the city center of Poland's capital, Warsaw, as an interactive 3D map made of voxels. Voxels are pixel-like blocks in three-dimensional space, and the project's visual style is clearly inspired by Minecraft and MagicaVoxel.
The author explains in the post that you can freely change the date, time of day, and weather, roam around the city, and even click on a pedestrian to take direct control of them — an interaction inspired by the classic game Dungeon Keeper. This "playability" sets it apart from traditional static 3D city models, making it feel more like a miniature, controllable city sandbox.
According to the author, the project is best experienced on desktop but has also been adapted for mobile where possible. The first load may be slow, which is related to the large amount of geometry that a voxel map needs to render.
Voxel is short for "volumetric pixel," referring to a cubic unit divided by a regular grid in three-dimensional space, similar to a pixel in a 2D image. Each voxel can carry attributes such as color and material, and a large number of voxels combined can express complex 3D shapes. Unlike traditional polygon mesh modeling, voxel models are inherently discrete and blocky in style, and rendering typically involves handling a large number of cube instances, placing high demands on the organization and optimization of geometric data. Voxels are widely used in games, medical imaging, and scientific visualization, while tools like Minecraft and MagicaVoxel have brought this style to the masses.
Where the Data Comes From: Three Layers of Open Datasets
The most noteworthy aspect of this project is that it is built entirely on public data. The author lists three main data sources:
OpenStreetMap: The City's Skeleton
OSM provides basic elements such as buildings, streets, sidewalks, and tram tracks. These vector data form the geometric skeleton of the voxel city, determining where buildings, roads, and tracks are.
OpenStreetMap (OSM) is an open map database maintained collaboratively by volunteers worldwide, released under an open license that allows free use, modification, and distribution. OSM data is organized as vector features, including nodes, ways, and relations, and can express rich information such as building footprints, road networks, land use, and points of interest. Because the data is contributed by the community, its coverage and accuracy vary by region, but it is usually quite complete in European cities. For developers, OSM is not just a base map but a programmable geographic data source, often used for routing, urban analysis, and 3D city modeling.
GUGiK / Geoportal: Height and Color
The national geoportal of GUGiK, Poland's surveying and geographic information agency, provides LiDAR data used to estimate building heights and tree heights, as well as aerial imagery used to extract colors for buildings and the ground. This explains why the voxel buildings are not uniformly gray-white but display relatively realistic tones.
LiDAR (Light Detection and Ranging) is an active remote sensing technology that calculates the distance to surface objects by emitting laser pulses and measuring reflection times, thereby generating high-precision 3D point cloud data. In urban construction, airborne LiDAR can penetrate gaps in vegetation to obtain elevation information for building rooftops and the ground, from which building heights and tree canopy heights can be estimated. Aerial imagery refers to high-resolution orthophotos captured from aircraft or drones, commonly used to extract surface colors and textures. Combining LiDAR's height information with aerial imagery's color information provides relatively realistic vertical profiles and tones for the voxel city.
Warsaw Urban Tree Database: Seasonal Details
Warsaw's municipal map service records the tree species of individual trees. The author uses this data to make vegetation of different tree species behave differently as seasons change. This is a remarkably delicate design — seasons are not just an overall color shift but are tied to real tree species.
Open Source and Reusable: Another Layer of Value
The project is released as open source on GitHub under the repository name stared/warsaw-vibe-voxel. The author explicitly invites others to freely (re)use it. This means the pipeline of "open data to voxel city" has the potential to be reused for other cities.
From a technical workflow perspective, the project demonstrates a feasible path: vector map data provides the planar layout, LiDAR provides vertical heights, aerial imagery provides colors, and the tree database provides ecological details, all ultimately consolidated into a voxel representation. Any city with similar open data could theoretically follow this path.
The Significance of Voxel Cities: Lightweight, Playable, Accessible
Traditional 3D urban visualization often relies on high-precision modeling or oblique photogrammetry, which involves large data volumes and high production costs, making it difficult for ordinary people to participate. Voxelization is a deliberate reduction in dimensionality: it sacrifices photorealism in exchange for a smaller data footprint, stronger stylized expression, and a lower barrier to interaction.
The value of such projects is not just that they "look good." They stitch together urban planning data, geographic information, and gamified interaction, allowing the public to understand the city they live in in an intuitive and fun way. Designs like changing time and weather or clicking on pedestrians also turn data visualization from something you "look at" into something you "play with."
Of course, the project still has obvious limitations: slow initial loading, limited mobile experience, and a voxel scale that means it cannot show architectural details. But as an open-source experiment based on open data, it provides a clear example — a city can be more than lines on a map; it can also be a blocky world you can walk into.
Oblique photogrammetry is a technique that captures the same area from multiple angles and uses photogrammetric algorithms to reconstruct a 3D model, generating dense meshes or point clouds with realistic textures. It is widely used in real-scene 3D city modeling, but the data volume is enormous, placing high demands on storage, transmission, and real-time rendering, and post-processing is complex. In contrast, voxelization is a deliberate abstraction and dimensionality reduction that replaces continuous surfaces with regular blocks, greatly reducing geometric complexity while preserving the city's basic spatial relationships, making it better suited for lightweight, interactive web applications.
Related articles
How Accurate Are Sci-Fi's Predictions About AI?
How accurate are sci-fi's predictions about AI? Drawing on a Hacker News discussion, this article analyzes sci-fi's foresight and blind spots in AI development, arguing its true value lies in inspiring thought rather than precise prediction.
SwarmEvolve: Latent-Space Agent Swarms Accelerate Evolutionary Search
SwarmEvolve lets hundreds of low-cost agents swarm-reason in latent space, sharing info via KV caches to achieve fast evolutionary search for math and science problems—reaching in a median of 3–8 minutes the best scores other frameworks need 60 minutes for.
trainlint: Static Checks Catch ML Training Bugs That Don't Crash
trainlint is a static analysis tool that finds ML training bugs that don't throw errors, including data leakage, leftover eval mode, and softmax/CrossEntropyLoss conflicts across 14 categories. This article covers its principles, validation results, and limitations.