Many building generation approaches focus on functional requirements or life-size concerns, creating a gap for aesthetic-focused building generation systems. We present a system that generates utopian mega-structures using procedural content generation techniques. Our goal is that the output of the system can serve as inspiration and visual explorations, or be materialized into dioramas, puzzles, scenographic elements, or graphic design material. Buildings can be much more than merely functional spaces. Fictional genres such as sci-fi and cyberpunk have presented audiences with ever-growing, busy buildings of questionable functionality that strive for visual appeal and creativity, while also speculating about what the future of architecture might look like. The idea for this system stems from these utopian notions, forgoing functionality to promote creative believability. The building generation system presented in this pictorial employs several Procedural Content Generation (PCG) techniques common in Procedural Generation of Buildings (PGB) [1], such as Generative Grammars (GG) [2, 3] and Pseudo-Random Number Generation (PRNG) [4, 5], and generates complete, non-furnished building models.
To make the most out of PRNG and GG techniques, we felt it would be useful to define and analyze a
target set of outputs from which we could abstract similarities, hierarchies, and variations. The main
architectural inspiration for this project was the work of Ana Aragão [
Given our set of target outputs, we opted for a common structure for all buildings: a series of vertically disposed sections, comprised of several towers, whose walls are populated with modules (see Figure 1). The building generation is controlled by the following set of parameters: • Number of Sections; • Minimum and Maximum Tower Height; • Filler Height Tolerance; • Section Contraction and Expansion Limit; • Window Minimum and Maximum Scaling;
First, the section bases are partitioned by a simple generative grammar. All terminal sections are then extruded upward by a random height factor between MinimumTowerHeight and MaximumTowerHeight to form towers. Having separated towers allows the system to individually customize and populate diferent towers or parts of a section with diferent properties, not unlike what we see in our intended outputs.
Given the height diferences between towers, simply placing the next foundation over the tallest tower would not look believable; therefore, the towers are then re-extruded if they are within a certain height threshold, defined by FillerHeightTolerance (see Figure 2). The bounding box of the topmost areas is then scaled up or down by a factor between SectionContractionLimit and SectionExpansionLimit, and becomes the starting surface for the next section. This process is repeated a number of times equal to NumberOfSections.
Once all sections are generated, the system populates the visible walls with any number of userdefined modules. First, the system filters out any faces or parts of faces that are not visible from the outside. The height and width of each resulting face defines the upper and lower limits for the number of rows and columns of a grid, and a 3D Simplex noise space selects the specific module to be placed at each grid point, resized proportionately to the wall’s dimensions. In the future, we hope to accomplish the module application step using GG techniques, rather than a noise space. The source code repository can be found at: https://github.com/GuilhermeAlmeidaG8/LDC-Reclaimers.
Our findings demonstrate that the system can create a wide variety of buildings, both with diferent parameter sets and within each set (see Figures 4, 5 and ??). To this efect, we will present two representative outputs for a few diferent parameter sets, highlighting the most relevant ones. The parameters will be written in the figure captions, following this format:
The results, although preliminary, already demonstrate the system’s possible applications. The final
models can be 3D printed to make scenographic elements at a variety of scales or be used as decorative
pieces (see Figure 6). Compositions of multiple buildings with varying styles can be created based
on their arrangement, and the buildings’ components can be separated into physical assembly kits or
puzzles, similar to Santos et al. [
This work is funded by national funds through FCT – Foundation for Science and Technology, I.P., within the scope of the research unit UID/00326 - Centre for Informatics and Systems of the University of Coimbra.
During the preparation of this work, the authors used ChatGPT to: Paraphrase and reword, improve writing style. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the publication’s content. Figure 7: Vectorized render of a set of buildings generated by the system.