Architectural alchemy: Leveraging Artificial Intelligence for inspired design – a comprehensive study of creativity, control, and collaboration

Lenka Petráková, Vladimír Šimkovič

Cite this article
Petráková, L., Šimkovič, V. (2023) ‘Architectural alchemy: Leveraging Artificial Intelligence for inspired design – a comprehensive study of creativity, control, and collaboration’, Architecture Papers of the Faculty of Architecture and Design STU, 28(4), pp. 3-14.



Traditional architectural design provides a human-centric and intuitive way of creating spaces based on personal creativity, experience and cultural history. The final design usually unfolds linearly with predetermined approaches set in early stages, and sometimes, the experience can overshadow objective evaluation. In contrast, the design approach enhanced by Artificial Intelligence (AI) leverages technology to expand the boundaries of inspiration and creativity, offering new ways for exploration and innovation. AI enhances creativity by allowing architects to experiment with novel forms, structures, and ideas at an unprecedented pace. This fosters an environment where architects can explore and innovate more freely and faster without the typical constraints of manual design.

This research paper explores the complex role that Artificial Intelligence can occupy in architectural design. Contrary to the view of AI as a mere utility tool, the paper posits that AI can function as a collaborative partner, advancing human creativity by offering innovative design possibilities. Originating from 1950s computer science explorations, AI has come a long way to permeate various industries, including architecture, where it is notably propelled by recent advances in machine learning algorithms like Generative Adversarial Networks (GANs).

This paper delves into the effectiveness of AI-driven design approaches, exploring new ways of inspiration and innovation in the architectural sector while researching how we can control AI in the design process and use it as a tool instead of an autonomous designer. When working with the two platforms (Midjourney AI and Stable Diffusion), the questions are multifaceted and require careful consideration: How do text-to-image and image-to-image generation algorithms contribute to a more vivid visualisation of designs? How can we enable greater control and flexibility in the design process? What are their comparative strengths and limitations in the context of architectural design? How can AI’s role be moderated within the design process to ensure it functions as a collaboratively interactive tool rather than an autonomous designer? By focusing on these questions, the paper aims to investigate the mechanics of these platforms, evaluating their relative capabilities and providing insights into how they can be effectively harnessed in modern design practices.

To empirically validate these concepts, our paper conducts a comprehensive three-phase investigation featuring nine tests that meticulously assess the strengths and shortcomings of two leading AI platforms: Midjourney AI and Stable Diffusion. These platforms harmonise human creativity with AI-generated solutions by utilising features such as text prompts and image references, and they open up unprecedented avenues for innovation in architectural design.

Our comparative analysis shows that Midjourney AI excels in creating initial design concepts based on text prompts, mainly due to its extensive data libraries. However, it is deficient in refining these designs and providing designers with adequate control. Conversely, Stable Diffusion offers greater control to designers through features like ControlNet, enabling the selection of various control mechanisms. Nevertheless, Stable Diffusion’s generated visuals may lack definition compared to Midjourney AI, mainly because its generative models are smaller. In both systems, a standard limitation is an emphasis on shape and aesthetics at the expense of understanding the functionality of the given geometry.

Building on our empirical findings, the paper illustrates how designers can exert nuanced control over this emerging AI-driven design methodology to optimise workflow. The tests we conducted provide invaluable insights into these AI platforms’ capabilities and limitations. They also offer practical guidelines for overcoming these challenges through a balanced, hybrid approach that amalgamates the best elements of Midjourney AI and Stable Diffusion. Instead of positioning AI as a rival to human ingenuity, our methodology envisages it as a valuable adjunct, enhancing the collaborative potential between humans and machines in architectural design.

The research validates two key hypotheses regarding the harmony of creativity, control, and collaboration, stressing that human architects and AI platforms benefit from iterative feedback and ongoing adaptation. In conclusion, the study asserts that AI is not just a technological supplement but a transformative catalyst that has the potential to redefine the architectural design process fundamentally. It further emphasises that while AI can amplify and extend human creative instincts, the essence of creativity remains a uniquely human attribute. As such, the paper foresees a future for architectural practice that is both technologically advanced and artistically profound, thereby heralding a new paradigm in which human expertise and machine capabilities coalesce to create enriched design outcomes.

In summary, this paper contributes significantly to the ongoing conversation about integrating AI and machine learning in architectural design. The paper advocates for a balanced, dynamic partnership between human creativity and technological innovation, explaining the transformative potential inherent in such collaborations.

Keywords: AI, architecture, design, creativity, artificial intelligence