Abstract
Generative models for architectural design can have difficulty adhering to complex constraints, such as directional rules from specific cultural frameworks, which may result in layouts with functional or symbolic limitations. A computational framework can address this by integrating rule-based logic with generative models. The system can translate user requirements and geospatial data into an abstract graph where rooms are represented as nodes. A logic engine, which can utilize a component such as a graph neural network, can optimize the placement of these nodes according to predefined adjacency and directional rules. The resulting optimized spatial graph can be used to create a conditioning mask that guides a generative model, such as a latent diffusion model, to render a floor plan. This approach can facilitate the creation of architectural designs that are more likely to comply with specified functional and symbolic orientation rules, potentially reducing a need for subsequent manual correction.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Goyal, Arpit, "Architectural Design Generation Using Graph-Based Constraint Optimization and Conditioned Generative Models", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/9426