Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automating Computational Design with Generative AI

Joern Ploennigs, Markus Berger

TL;DR

This study evaluates diffusion-model-based generative AI for automated floor-plan design in civil engineering, identifying limitations in out-of-the-box use and proposing four semantic-encoding finetuning schemes (R, SR, SE, SRE). Through procedurally generated training data and textual-inversion fine-tuning on Stable Diffusion v2.1, the authors demonstrate that semantic encoding—especially color-encoded symbol semantics (SE)—significantly increases plan validity up to 90% in their tests. They also reveal persistent challenges in counting, symbol semantics, and translating generated outputs into BIM-ready representations, underscoring the need for BIM-aware diffusion models. The paper outlines a broader future direction including image-based diffusion enhancements and BIM-based diffusion models leveraging graph representations to enable direct 3D BIM generation and scalable data ecosystems. Overall, the work provides a concrete evaluation framework, demonstrates substantial gains from domain-specific fine-tuning, and maps clear paths toward fully automated, BIM-integrated computational design.

Abstract

AI image generators based on diffusion models have recently garnered attention for their capability to create images from simple text prompts. However, for practical use in civil engineering they need to be able to create specific construction plans for given constraints. This paper investigates the potential of current AI generators in addressing such challenges, specifically for the creation of simple floor plans. We explain how the underlying diffusion-models work and propose novel refinement approaches to improve semantic encoding and generation quality. In several experiments we show that we can improve validity of generated floor plans from 6% to 90%. Based on these results we derive future research challenges considering building information modelling. With this we provide: (i) evaluation of current generative AIs; (ii) propose improved refinement approaches; (iii) evaluate them on various examples; (iv) derive future directions for diffusion models in civil engineering.

Automating Computational Design with Generative AI

TL;DR

This study evaluates diffusion-model-based generative AI for automated floor-plan design in civil engineering, identifying limitations in out-of-the-box use and proposing four semantic-encoding finetuning schemes (R, SR, SE, SRE). Through procedurally generated training data and textual-inversion fine-tuning on Stable Diffusion v2.1, the authors demonstrate that semantic encoding—especially color-encoded symbol semantics (SE)—significantly increases plan validity up to 90% in their tests. They also reveal persistent challenges in counting, symbol semantics, and translating generated outputs into BIM-ready representations, underscoring the need for BIM-aware diffusion models. The paper outlines a broader future direction including image-based diffusion enhancements and BIM-based diffusion models leveraging graph representations to enable direct 3D BIM generation and scalable data ecosystems. Overall, the work provides a concrete evaluation framework, demonstrates substantial gains from domain-specific fine-tuning, and maps clear paths toward fully automated, BIM-integrated computational design.

Abstract

AI image generators based on diffusion models have recently garnered attention for their capability to create images from simple text prompts. However, for practical use in civil engineering they need to be able to create specific construction plans for given constraints. This paper investigates the potential of current AI generators in addressing such challenges, specifically for the creation of simple floor plans. We explain how the underlying diffusion-models work and propose novel refinement approaches to improve semantic encoding and generation quality. In several experiments we show that we can improve validity of generated floor plans from 6% to 90%. Based on these results we derive future research challenges considering building information modelling. With this we provide: (i) evaluation of current generative AIs; (ii) propose improved refinement approaches; (iii) evaluate them on various examples; (iv) derive future directions for diffusion models in civil engineering.
Paper Structure (26 sections, 2 equations, 11 figures, 1 table, 1 algorithm)

This paper contains 26 sections, 2 equations, 11 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. State of the Art in Computational Design
  3. Introduction to Diffusion models
  4. Model Structure
  5. Fine-tuning Diffusion Models
  6. Methodologies for Generative Design
  7. Idealized Integrated Workflow
  8. Existing Out-of-the-Box Workflow (Model $\mathcal{B}$)
  9. Proposed Fine-tuning Approach (Model $\mathcal{R}$)
  10. Proposed Improved Fine-Tuning with Color Encoding of Rooms (Model $\mathcal{SR}$) and Elements (Model $\mathcal{SE}$)
  11. Model Training
  12. Generation of the Training Data
  13. Structure of the Training Data
  14. Fine-Tuning the Models
  15. Evaluation
  16. ...and 11 more sections

Figures (11)

  • Figure 1: Examples of iterative denoising steps for a building floor plan in Midjourney v4.
  • Figure 2: Idealized integrated workflow for AI-based design
  • Figure 3: Generation Workflow for out-of-the-box Diffusion Model
  • Figure 4: Example floor plan of a house with garden created by the out-of-the-box model $\mathcal{B}$.
  • Figure 5: Workflow for a refined 2D/3D bitmap-based Diffusion Model
  • ...and 6 more figures