ARCHITECTURE AND AI

words by Zuzana Grúňová

Is architecture an art or is it a technical discipline? Is it only art when it borders on the unfeasible in the physical world? Conversely, if we prioritize the technical view and reduce the design to the most effective structures, materials, ergonomics, and typology, regardless of the aesthetic impression – is it still architecture?

My experience and emotions, tied to my entire career as an architect, say that architecture is both and everything. It can range from one pole to another. It can be visionary art that inspires and raises new questions, even if it may not be feasible. The projects and designs of architectural celebrity Zaha Hadid were exactly such a start from abstraction to bold forms of realized, built architecture.

Optimization to the minimum, even that has been here before. Ornament is a crime, less is more… but, I admit, I have never perceived any architecture of modern style, functionalism as truly cleansed of any ornament. In Loos’s buildings, the ornament was created by careful detail and texture, the structure of the material, and a precisely assembled composition of volumes and lines. Architecture, if it is to be of high quality and complete, probably cannot be without aesthetics, harmony.

Architecture is somehow straddled between art and construction, between feelings, colors, shapes, composition on one bank, and techniques, possibilities of materials, technologies, standards, safety, and utility properties on the other. The creator flows between them like a river – some “spill out” only on the bank of art, others create very utilitarian products, often mass-built on the technical other bank.

Try asking someone what building they remember – I’ve heard answers from acquaintances like – ‘Gaudí, any Gaudí, those colors and shapes are amazing!’ and also ‘Zápoľský Chapel in Spiš in Slovakia, that silence… and then the sun shone through the stained glass. Gothic is incredible.’ Or ‘I was transferring at a large transport hub in France, Liège-Guillemins. I don’t know if it was built by a famous architect, but I really liked that white lightness. A huge hall, but it looked like a feather.’ We may not know Santiago Calatrava, but we can certainly feel the emotion that his architecture brought.

My experience with AI directly in art and architecture is not extensive. They are rather the attempts of an enthusiast and explorer, not a systematic survey or even the development of improvements. I am learning and exploring.

Where would I see the role and help of AI in architectural design? I will now focus exclusively on the first, perhaps the most artistic, most emotional first step – creating a distinctive, attractive design, an architectural study. Study explores possibilities and creates the first impression in the dialogue between the architect/interior designer and the investor.

At the beginning of every building is the initial design. This is shaped by the investor’s requirements (more or less specific) regarding function and parameters (number of rooms, beds, offices, storage area, connections and algorithms of processes that the building should contain, and so on). On the other hand, there is the land – orientation, terrain, climate, urbanism, greenery, and hundreds of other inputs.

AI could significantly help with generative design here. In collaboration with a human creator, it could sort the importance of inputs into a structure of priorities. Is the view and connection with the city more important? Or is the most important issue the correct orientation of the facade so that it does not unnecessarily overheat in the summer? It would generate more variants, each with its own profile of parameters. Is the building more energy-efficient, or does it better integrate the production process that is to take place in it? Is it visually interesting, but what influenced that?

AI should be able to very effectively and interactively manage fine optimization to the desired result and process it understandably in the form of an interactive model, video – walks around the building and its interior so that it is possible to assess the resulting volumes and shapes from the perspective of a person, user, inhabitant of the building. It could also help create visually attractive presentations of the design and thus increase the competitiveness of the creator. It should provide an advantage in the market – saving costs, energy and time and at the same time the interactivity of responses to requirements.

What is the current state and possibilities? Overviews state that so far, generation is mainly available using tools with the text-to-image (txt2img) feature, i.e., creating 2D images based on a text input, prompt. This can be further modified and fine-tuned by creating variants, using styles.

The second basic option is image-to-image (img2img), i.e., working with image editing. For example, creating a variant from it, using what is called inpainting or outpainting. Inpainting is replacing part of the image by erasing an unsuitable part and inserting another (a new table into the scene). Outpainting/uncropping mean adding part of the scene beyond the original image, generating the scene beyond the edge of the original one. A beautiful example of a workflow from generating the initial image to the final edit, along with human intervention, is presented, for example, in the article by Ploenings and Berger on the use of AI in architecture. Their comparison of the possibilities offered by individual platforms is also useful, they compared DALL·E 2, Midjourney v4, and Stable Diffusion v2.1. The comparison clearly shows that, for example, for the field of architecture and interior design, we still lack sufficiently specific options of individual tools. There are gaps in the workflows, some very necessary tools are missing, or do not work ideally. [1]

Idea creation – this is exactly the purpose that txt2img tools master. It’s worse with the second step, which are 3D sketches. All tools were able to generate sketches in various styles, but an architectural bias in favor of better documented and more famous Western architecture is manifested. This is also mentioned by a series of articles by Nicolas Valencia about trends in architecture with the help of AI [2] or even trends in architecture predicted by AI [3]. Collage is the process of inserting new parts into the model, usually it can be done with inpainting. Generation using a prompt is less supported by tools. Similarly, not all tools support the combination of images of different buildings or surroundings/interior and their combination into a new composition. Building variants, such as inserting new parts of architecture (extensions, elements), are most often also realized as inpainting. According to the article, the generation of interiors works excellently. Also the generation of exteriors is amazing, especially in common scenarios (again architectural bias, predominance of a certain style and type of data). Textures and structures that architecture uses as part of its language supports to a greater extent only Midjourney. The Midjourney platform can create a 2D image “tile”[4, 5, 6], from which it is then possible to generate a pattern in other software and use it on a 2D or 3D model. [7]

In addition to the large predominance of certain data that are the source of generative AI models, there is also a bias in the input itself. 63.7% of websites are in English, so AI models are very likely to respond better and more accurately to prompts given in English. A certain filter, causing incomprehensibility, or worse results can also be caused by a language barrier of the creator, using non-native English.

However, none of the platforms and tools currently (summer 2024) understand the concept of converting 3D to 2D, i.e., generating 2D documentation. Floor plans, sections, and views are created by surfaces, lines as an abstract representation of real constructions. AI cannot understand the generated 2D display as a way of displaying part of a 3D object.

AI does not yet understand several specific architectural terms, for example, for the exact type of window or shape of the facade. The intervention of an architect with knowledge is necessary.

So far, we only have a 2D image available, which can be generated in a certain style, using the inspiration of the work of some (western) architect or architecture of a certain (western) country. We cannot use AI to transform this 2D image into a 3D model or create it directly as a 3D model, moreover interactive. The control of the extent to which the investor’s requirements and the location (realistically usable materials, safety, stability…) are reflected in the design is exclusively corrected and assessed by the human creator. This side still depends on the education, experience, and skills of the architect and his role is irreplaceable if the design is to be of high quality and feasible.

I’ll just add one more impression of mine, which was brought to me by researching various sources. On the second, “technical” river bank of construction, where architecture steps with the other foot, it is probably simpler. Most tasks are more precisely defined, basically it is about respecting the limits of various legislation or standards, creating an optimal solution based on a large amount of data and harmonizing requirements according to given priorities. A lot of data is numerical or it is about finding a defined element in the picture (for example, a crack, colouring of the structure).

AI can manage infrastructure intelligently, streamlining workflows, and optimizing resource utilization. It can create predictive models that can forecast the behaviour of materials and optimize processes. Project scheduling, resource allocation, risk assessment, and security management are many areas of construction management, that could be improved by using AI. AI-based geotechnical applications provide accurate estimation of soil properties, soil damage, and foundation construction. Transportation planning, traffic forecasting, intelligent transportation, and transportation improvements can be facilitated by AI applications. Image processing and continuous analysis huge amount of data  creates a role for AI in the monitoring and maintenance of public infrastructure, like bridge inspection, integrity assessment of pipelines and early detection of construction defects.[8, 9, 10]

Do you want to try some tools? Please, the selection is diverse! [11]

 


[1] Ploennigs, J., Berger, M. AI art in architecture. AI Civ. Eng. 2, 8 (2023). https://doi.org/10.1007/s43503-023-00018-y

[2] Nicolás Valencia. “The Impact of AI Tools on Architecture in 2024 (and Beyond)” 06 Dec 2023. ArchDaily. Accessed 31 May 2024. <https://www.archdaily.com/1009545/the-impact-of-ai-tools-on-architecture-in-2024-and-beyond> ISSN 0719-8884

[3] Nicolás Valencia. “An Artificial Intelligence Bot Picks The Architecture Trends of 2023” 21 Dec 2022. ArchDaily. Accessed 31 May 2024. https://www.archdaily.com/994020/the-architecture-trends-of-2023-according-to-chatgpt-an-ai-trained-model-bot> ISSN 0719-8884

[4] Ma, Y. (2023, July 3). Best Prompts for Midjourney’s –tile (seamless textures, patterns & maps). AITuts. <https://aituts.com/midjourney-tile/>

[5] K, M. (2023, August 13). Midjourney Prompts for Patterns (Tile Parameter). LetsTryAI. <https://letstryai.com/midjourney-prompts-for-patterns/>

[6] McMaster, W. [Wade McMaster]. (2022, September 30). MidJourney AI Art Tutorials and Videos Create Patterns with Midjourney! (Seamless AI Art Tiling) [Video]. YouTube. <https://www.youtube.com/watch?v=USHKJiQfq0E>

[7] Cheung, E. (n.d.). Seamless Texture Checker. PyCheung. Retrieved from https://pycheung.com/checker/

[8] https://www.statista.com/chart/26884/languages-on-the-internet/?utm_medium=website&utm_source=archdaily.com

[8] Li, H., Zhang, Y., Cao, Y., Zhao, J., & Zhao, Z. (2024). Applications of artificial intelligence in the AEC industry: a review and future outlook. Journal of Asian Architecture and Building Engineering, 1–17. https://doi.org/10.1080/13467581.2024.2343800

[9] HyScaler. (2024). AI in Civil Engineering: The Prominent Role of AI in AEC in 2024. Retrieved from https://hyscaler.com/insights/ai-in-civil-engineering/

[10] Chaos. (2024, April 23). The State of AI in Architecture: Insights from 1,200+ Architects & Designers. Architizer portal. https://blog.enscape3d.com/the-state-of-ai-in-architecture-survey-insights

[11] https://architizer.com/blog/practice/tools/top-ai-tools-for-architects-and-designers/

_TAGS

_PARTNERS

Skip to content