A Study on Exterior Design Alternatives for Temporary Residential Facilities Using Generative Artificial Intelligence

Abstract

The increasing frequency and severity of natural disasters—such as floods, storms, droughts, and earthquakes—have created a growing demand for temporary housing. These facilities must be rapidly deployed to provide safe, functional living environments for displaced individuals. This study proposes a design methodology for temporary housing exteriors using the text-to-image capabilities of generative artificial intelligence (GenAI) to address urgent post-disaster housing needs. The approach aims to improve both the efficiency and practicality of early-stage design processes. The study reviews global trends in temporary housing and the architectural applications of GenAI, identifying five key environmental factors that influence design: type of disaster, location and climate, duration of residence, materials and structure, and housing design. Based on these factors, hypothetical disaster scenarios were developed using ChatGPT, and corresponding exterior designs were generated using Stable Diffusion. The results show that diverse, scenario-specific design alternatives can be effectively produced using GenAI, demonstrating its potential as a valuable tool in architectural planning for disaster response. Expert evaluation of the generated designs confirmed their ability to adhere to text prompts but revealed a significant gap in terms of architectural plausibility and practical feasibility, highlighting the essential role of expert oversight. This study offers a foundation for expanding GenAI applications in emergency housing systems and supports the development of faster, more adaptable design solutions for communities affected by natural disasters.

1. Introduction

1.1. Background and Purpose of the Study

As the frequency and intensity of natural disasters—such as floods, storms, droughts, and earthquakes—continue to rise globally, the demand for temporary residential facilities is also rapidly increasing [1]. This trend is accelerated by factors such as climate change, population growth, urban densification in coastal areas, and inadequate disaster preparedness. In the aftermath of a disaster, displaced individuals depend on temporary housing until they can return to their original residences. These facilities play a crucial role in ensuring the survival and well-being of affected populations [2].

According to a report by the Centre for Research on the Epidemiology of Disasters (CRED), there were 399 disaster events in 2023, resulting in 86,473 deaths and affecting approximately 93.1 million people. The estimated economic losses reached USD 202.7 billion—surpassing the average over the past decade [3]. These statistics underscore the escalating frequency and scale of natural disasters and highlight the urgent need to strengthen disaster preparedness and response systems. Specifically, with the rise in climate-related disasters, designing effective temporary housing solutions has become essential [4].

In practice, two contrasting delivery logics shape how temporary housing is deployed at scale. Government-led systems such as FEMA [5] prioritize speed and standardization—e.g., trailers or prefabricated Temporary Housing Units (THUs) that can be rapidly mobilized [6]—but may be less adaptable to local climate, cultural context, or available materials. By contrast, international guidelines (e.g., UNHCR) emphasize cultural appropriateness, use of local resources, and climate- and site-specific adaptation to uphold dignity and longer-term viability, yet often face constraints in speed and scalability [7]. This inherent tension between rapid mass deployment and contextual quality is a core design dilemma in disaster relief architecture.

Addressing this challenge requires new workflows, as conventional design processes struggle to rapidly generate and adapt a diverse portfolio of context-specific alternatives. Concurrent with this need, the field of architectural design is undergoing a paradigm shift with the advent of generative artificial intelligence (GenAI) [8]. This technology has rapidly evolved from early models like Generative Adversarial Networks (GANs) to more sophisticated denoising diffusion probabilistic models (DDPMs), such as Stable Diffusion and Midjourney, which can generate high-fidelity images from textual descriptions [9,10,11]. Leading architectural firms are already integrating these tools into their conceptual design workflows to enhance creative ideation and accelerate visualization [12].

However, the application of GenAI in architecture, while promising, still lags behind advancements in computer science [13]. Much of the current use focuses on aesthetic exploration or general ideation for conventional building types. A significant opportunity remains to develop systematic, problem-solving methodologies that leverage GenAI to address specific and complex design challenges beyond simple image generation. Applying these advanced tools to the unique and critical constraints of emergency architecture—where resilience, rapid deployment, and material economy are paramount—remains a largely unexplored frontier.

In this context, this study proposes a methodology for swiftly generating exterior design alternatives for temporary residential facilities suited to different natural disaster scenarios. While considerable research has focused on the structural engineering of modular buildings for diverse applications [14,15], the conceptual front-end phase of the design—particularly the rapid, context-specific ideation required in disaster response—has received comparatively limited attention. To address this underexplored dimension, the study investigates the potential of T2I-based generative AI to provide a novel workflow that accelerates and diversifies design ideation at the earliest stages of architectural planning for emergency contexts.

Accordingly, this paper addresses the following primary research question: How can text-to-image (T2I) generative AI technology be utilized to rapidly generate diverse exterior design alternatives for temporary residential facilities tailored to various disaster scenarios? Furthermore, this study evaluates the practical applicability of the generated AI designs through a survey of architectural experts to assess their feasibility in real-world disaster response scenarios.

1.2. Scope and Methodology

In this study, a methodology capable of rapidly generating exterior design alternatives for temporary residential facilities was developed in response to the growing global incidence of natural disasters. By enabling the review of diverse design ideas and the comparison of visual alternatives during the early stages of design process, the proposed methodology can contribute to more effective design goals and directions. Figure 1 illustrates the overall research methodology.

Initially, prior studies were reviewed to analyze global trends in exterior design for temporary housing and the application of GenAI in architectural design, thereby clarifying the need for this research. Additionally, key environmental factors identified in domestic and international studies on temporary housing were examined to extract core prompt items for generating exterior design images using T2I technology. Based on this analysis, essential prompt items were defined to generate T2I-based temporary housing designs. Using these items, hypothetical disaster scenarios were created with the assistance of ChatGPT, and detailed prompts were developed accordingly. A Stable Diffusion-based T2I workflow was then constructed, through which exterior design alternatives were generated by inputting scenario-specific prompts. To validate the practicality of these alternatives, the methodology was extended to include an expert-based evaluation, where the generated designs were assessed through a survey of architectural professionals.

The study results not only contribute to enhancing the speed and efficiency of design efforts for temporary housing in disaster situations but also serve as a foundational reference for exploring broader applications of GenAI in the field of architecture. Thus, this study not only pursues a faster disaster-response housing design process using AI but also provides academic value by presenting a reproducible methodology and highlighting the role of GenAI in the field of architecture.

2. Literature Review

To establish the context for this study, this literature review synthesizes two primary research streams: the design of temporary residential facilities and the application of generative AI in architecture. The review is structured to first build a comprehensive understanding of the problem domain by analyzing established design trends in temporary housing, focusing on practical requirements (Section 2.1) and the growing importance of sustainability and resilience (Section 2.2). Subsequently, it examines the parallel evolution of the proposed technological solution, reviewing the foundational applications of GenAI in architectural design (Section 2.3) and its specific use in façade generation (Section 2.4). By systematically exploring these distinct but converging fields, this review identifies the critical research gap at their intersection: the lack of a methodology for applying advanced T2I technologies to the unique design challenges of emergency housing.

2.1. Trends in the Exterior Design of Temporary Residential Facilities

This section reviews previous research on the design of temporary residential facilities in both domestic and international contexts. Temporary housing offers rapid residential solutions in various emergency scenarios including natural disasters, refugee crises, and post-disaster recovery. Recently, design approaches emphasizing modularity, foldability, and sustainability have gained particular attention. Existing studies have focused on core aspects, such as assemblability, mobility, thermal insulation, and ecological viability.

Yu and Bai highlighted several issues in temporary housing, including single-space layouts, limited comfort, and poor ecological efficiency [16]. They proposed a modular and sustainable design framework, suggesting that factory production and foldable loading/unloading systems can enhance functionality and enable strategic stockpiling and reuse for greater ecological sustainability. Grakhov and Tolkachev analyzed modular architecture as an efficient means of providing rapid shelter in disaster-affected regions [17]. They systematized the design and installation processes of prefabricated buildings to improve the insulation, mobility, and mass assembly efficiency. Kang et al. proposed a folding-based design for temporary housing that emphasized mobility, storage efficiency, and spatial adaptability [18]. They developed a prototype to address common issues in existing facilities—such as poor mobility, a lack of privacy, and inadequate insulation—by employing foldable structures for rapid assembly and efficient deployment. Kang and Kim highlighted the benefits of dome structures in terms of their assembly and transportation, demonstrating that dome structures offer better insulation and energy efficiency than traditional container-based systems, which makes them suitable for medium- to long-term residences and rapid deployment in emergencies [19]. Moran et al. applied integrated transdisciplinary tools (ITDT) to address the complex engineering challenges in refugee housing design [20]. They incorporated methods such as Kano analysis for user requirements, teoriya resheniya izobretatelskikh zadach (TRIZ) for inventive problem solving, and interpretive structural modeling (ISM) for structuring system components, effectively eliminating design conflicts and generating innovative alternatives. Guzhova and Khairullin conducted a comparative analysis of containers and prefabricated structures, emphasizing the benefits of modular designs [21]. They concluded that prefabricated systems outperform conventional constructions in terms of insulation and mobility and are broadly applicable to various disaster scenarios.

Collectively, these studies establish the core engineering and logistical requirements for effective temporary housing, such as modularity and mobility. However, their primary focus remains on physical performance and deployment, with limited attention given to the conceptual front-end: the rapid generation of diverse, context-aware design alternatives tailored to specific disaster scenarios. This study addresses that gap by employing GenAI to explore a more agile approach to emergency housing ideation.

2.2. Trends in Temporary Housing Design from the Perspective of Sustainability and Resilience

A growing body of research has emphasized the importance of sustainability and resilience in the design of temporary housing used in the aftermath of natural disasters. Montalbano and Santi analyzed global disaster case studies and identified environmental, economic, and social issues associated with current temporary housing units (THUs), proposing essential design requirements to address these challenges [22]. Through case analysis and requirements definition, they proposed design strategies that balanced cost-effectiveness, social acceptability, and environmental impact. The study also highlights the need to enhance sustainability by incorporating digital tools such as building information modeling (BIM) and circular economy principles into future temporary housing solutions. Similarly, a comprehensive literature review by Perrucci and Baroud revealed the need for further research in areas such as housing typology, site selection, and cost optimization [2]. Their findings pointed to insufficient efforts in pre-disaster planning, large-scale storage, sustainable design, and community resilience and recommended the application of established frameworks—such as Leadership in Energy and Environmental Design (LEED) or Sheltering and Temporary Essential Power (STEP)—to improve energy efficiency and reusability. They also advocated the integration of circular economy concepts to facilitate the recycling of THUs and reduce environmental burdens.

Resilience to climate change has become a crucial consideration in temporary housing design. Ruíz and Mack-Vergara compared resilient and sustainable housing models in urban contexts [23]. They found that resilient housing requires secure site selection, robust structural materials (e.g., concrete), and access to alternative water and energy sources. The focus of sustainable housing is climate-adapted design, efficient use of energy and water, high indoor environmental quality, and the use of locally sourced materials. Both models prioritize the use of durable materials, implying that even temporary shelters must be designed with long-term resilience and resource autonomy in mind. To further enhance sustainability, digital tools and automation are increasingly being adopted into design methodologies. Tirella et al. proposed a BIM-based modular housing platform that enables the automated configuration of temporary housing, even by non-experts [24]. The system allows users to input site conditions and housing needs, automatically generating optimized eco-friendly exterior designs through the integration of location-specific energy assessments and life-cycle carbon footprint analyses. This approach illustrates the dual aim of achieving both speed and sustainability in temporary housing design. Finally, Friedman emphasized the role of circular economy principles in architectural design, highlighting strategies such as design for disassembly (DfD), modular fabrication, and reuse-oriented design to maximize sustainability [25]. While these approaches promote resource efficiency and waste reduction, he also identified real-world barriers such as storage and logistics challenges, high initial costs, regulatory gaps, and limited market demand. Prior research on sustainable temporary housing provides a crucial framework of design principles, emphasizing energy efficiency, material circularity, and adaptability. These studies effectively define the goals for next-generation emergency shelters. However, they do not address the procedural challenge of how to rapidly generate and compare a multitude of design options that adhere to these complex principles in the time-critical early stages of a post-disaster response. A key limitation, therefore, is the absence of a workflow that connects these sustainability goals with agile, generative design tools.

2.3. Trends in GenAI Applications in Architectural Design

This section reviews domestic and international studies on the application of GenAI in architectural design. GenAI is now widely adopted across various creative industries—including art, fashion, design, advertising, game development, and film production—and its potential in architecture is rapidly expanding.

Lee and Ko explored the feasibility of applying AI-based T2I generators in architecture, introducing the foundational concepts for integrating GenAI into the architectural design process [26]. Li et al. proposed a novel architectural workflow incorporating GenAI, enabling the generation of conceptual floor plans and 3D models from simple sketches, along with the rapid, controlled production of architectural renderings from textual descriptions [6]. Tan and Luhrs examined how T2I tools such as Midjourney can function as creativity-enhancing instruments in architectural design, highlighting the potential for blending creative ideation with technical execution [27]. Jo et al. presented an approach that uses GenAI to generate architectural alternatives reflecting regional identity, exploring how localized features can be rapidly integrated into building designs [28]. Lee et al. proposed a new method for enhancing architectural visualization using GenAI, particularly T2I models [29]. Their study generated over 10,000 housing model images reflecting the stylistic preferences of individual architects, thus demonstrating the utility of AI in improving design visualization efficiency.

These studies demonstrate the growing potential of GenAI as a powerful tool for accelerating ideation and visualization in standard architectural workflows. However, their application remains largely confined to general building typologies and stylistic exploration. A significant gap persists in applying these technologies to specialized, high-stakes fields like disaster response, where the design challenge is not just aesthetic but is fundamentally about speed, functionality, and context-specific adaptation. This research aims to bridge this gap by developing a systematic methodology for this specific use case.

2.4. Trends in Applications of GenAI and T2I Technologies in Architectural Façade Design

Recently, architectural design has witnessed increasing interest in GenAI applications, particularly T2I techniques, for the automatic generation of design alternatives. Li et al. reviewed the application of AI in architectural design and emphasized the potential of natural-language-based generative technologies to significantly enhance designers’ creative processes [30]. They explained that large language models (LLMs) based on transformer architectures can interpret textual design requirements and link them to image-generation models (e.g., Midjourney, Stable Diffusion), enabling the rapid visualization of designer intent. For practical implementation, Yang and Qian proposed a T2I-based framework for conceptual high-rise design [31]. They developed a two-stage generative model that interprets textual design descriptions, extracts abstract architectural concepts, and generates corresponding architectural images. In the first stage, architectural concepts are identified using a contrastive learning-based text encoder, and are then transformed into conceptual sketches. In the second stage, a pretrained diffusion model converts these sketches into photorealistic renderings of building facades. Their findings demonstrated that plausible high-rise design alternatives could be automatically produced from text prompts alone. A comparative evaluation using real-world cases validated the effectiveness of their approach. These studies highlight the growing value of natural-language-driven tools in enhancing early-stage design efficiency.

Research specifically targeting GenAI applications in architectural façade design has also emerged. Ali and Lee developed the iFACADE system, which blends the styles of two adjacent buildings to generate context-appropriate facades for urban infill projects [32]. By extracting geometric and proportional features (e.g., building form and window-to-wall ratio) from the surrounding architecture, the system uses a generative adversarial network (GAN)-based model to synthesize new facade designs that harmonize with the local environment. Their prototype demonstrated the potential of such systems as communication tools between architects and clients during early design phases. Generative deep learning has also been applied in renovation contexts. Lin and Song used GANs for the façade remodeling of aging industrial buildings to improve urban regeneration efficiency by automatically generating diverse design alternatives [33]. In their study, various transformation techniques were applied to the original façade images to construct a dataset that mapped multiple potential remodeling outcomes. This dataset was then used to train the GAN, enabling the generation of facade improvement plans that balance feasibility and creativity. Their study demonstrated that AI can rapidly propose diverse design alternatives for many factory buildings, significantly expanding the range of options available to designers. This illustrates that GenAI can serve not only in the creation of new architecture but also as a valuable tool for exploring design alternatives for the revitalization of existing architectural heritage.

Notably, technical advances have been made in generating comprehensive building facades. Wan et al. studied a deep learning method for simultaneously generating all five sides of a low-rise house (four elevations and the roof) [34]. While most existing models focus only on the front façade, they built a Pix2Pix-based generator trained on Solar Decathlon competition data. To enhance performance, they compared variations in the U-Net architecture, including U-Net++, HRNet, and Attention U-Net, and found that the attention-based model delivered the highest-quality results. Their work demonstrated that AI can holistically propose building envelopes, potentially reducing the time required for detailed housing designs. Although additional studies are emerging—such as those integrating blockchain into AI-driven design workflows or extending T2I techniques into video-based architectural visualizations—empirical comparative research on GenAI platforms in practical architectural settings is still in its early stages. Nonetheless, The current literature confirms that GenAI is a valuable tool for rapidly producing design alternatives for architectural façades. However, these applications are predominantly focused on static, urban environments where aesthetic harmonization is the primary driver. A critical limitation is their lack of relevance to dynamic, scenario-based challenges, such as designing temporary housing where functional resilience and rapid deployment are more critical than stylistic novelty. This research, therefore, pivots from using GenAI in static contexts to applying it as a problem-solving tool for urgent, variable disaster scenarios.

2.5. Summary

The distinctiveness and originality of this study stem from its integration of practical demands in temporary housing design with recent advancements in architectural design technologies. As discussed in Section 2.1, the core design considerations for temporary housing in various natural disaster contexts can be summarized as assembly, mobility, thermal insulation, and cultural appropriateness, with particular emphasis on rapid deployability through modular and foldable structures. However, previous studies have generally addressed the engineering and logistical challenges of deployment rather than the rapid generation of context-specific conceptual designs.

As discussed in Section 2.2, sustainability and resilience are becoming increasingly important in temporary housing designs. There is a growing need for integrated design frameworks that address environmental compatibility, material recyclability, and energy efficiency.

Section 2.3 presents a review of the growing applications of GenAI, particularly T2I technologies, as a visualization tool in architectural design. These technologies have proven effective in the early stages of design by enhancing idea generation and visual expression. Nonetheless, most related studies have focused on general housing typologies or artistic imagery, with limited applicability to the urgent and context-specific requirements of temporary housing in disaster situations.

Section 2.4 examines the potential of GenAI to automate façade design; however, existing research has primarily addressed design generation within static contexts, with limited attention to dynamic, scenario-based applications such as disaster response.

Addressing these limitations, this study proposes a new design workflow that employs T2I-based GenAI to model a range of natural disaster scenarios. This workflow facilitates the rapid generation of exterior design alternatives that reflect specific environmental conditions, material characteristics, occupancy durations, and housing types. By quantifying the tailored prompts for each disaster scenario, this study verified the practical applicability of GenAI for visualizing designer intent within urgent design processes. This methodology helps overcome the repetitiveness and rigidity often associated with conventional temporary housing designs and significantly enhances the diversity and speed of design responses required during actual disaster situations.

Building on the foregoing synthesis, this review clarifies how the sustainability and resilience requirements of temporary housing (Section 2.1 and Section 2.2) intersect with recent advances in multi-scenario, AI-driven generative design (Section 2.3 and Section 2.4), and it positions our study’s practical contribution as a workflow that rapidly produces exterior alternatives tailored to disaster-specific conditions. In doing so, the literature converges on three actionable gaps that motivate the method developed in the following sections: first, a methodological gap—a lack of a systematic, time-critical workflow for generating and comparing diverse, context-specific options in emergency architecture; second, an application gap—the predominant focus of architectural AI on conventional building types rather than function-driven disaster response; and third, an integration gap—the insufficient translation of real-world constraints of temporary housing into effective guidance for generative tools. The proposed workflow is designed to address these three gaps in concert.

3. Prompt Items and Scenario Development for Generating Exterior Design Alternatives for Temporary Residential Facilities

3.1. Analysis of Environmental Factors in Domestic and International Temporary Housing Cases

In disaster contexts, the design of temporary residential facilities plays a critical role beyond merely providing shelter; it also ensures the survival and stability of displaced individuals. Therefore, it is essential that such designs consider the geographical and climatic conditions, cultural context, and disaster types specific to each affected region. This study identifies ten key environmental factors relevant to the design of temporary housing and analyzes five major references (labeled A–E) based on these factors. The core focus and contributions of each reference are summarized below, as shown in

Table 1. Analysis of environmental characteristics in temporary residential facilities.

No.	Factor	Description	A	B	C	D	E
1	Economic Feasibility	GDP level of the disaster-affected area and availability of economic support	◎	〇	△	〇	◎
2	Location	Location of temporary housing (urban, rural, etc.) and natural environmental conditions	◎	〇	△	◎	◎
3	Housing Duration	Duration of housing needs: short-term, mid-term, or long-term	〇	◎	△	〇	◎
4	Structural Stability	Durability, prefabricated structure type, and resistance to environmental factors	△	〇	◎	〇	◎
5	Type	Housing types such as standalone, collective, prefabricated, or mobile units	〇	◎	〇	△	◎
6	Materials	Use of local materials (wood, metal, plastic, etc.) and recyclability	〇	◎	〇	〇	〇
7	Environmental Suitability	Climatic conditions (temperature, rainfall, etc.) and harmony with the local environment	△	◎	〇	〇	〇
8	Hygiene and Health	Availability of sanitary facilities (toilets, water supply, etc.) and disease prevention	◎	〇	◎	〇	△
9	Social Sustainability	Interaction with local residents and potential for long-term use	〇	△	〇	◎	◎
10	Cultural Considerations	Incorporation of local cultural and religious characteristics in design	△	〇	△	◎	〇

Note: ◎, 〇, and △ indicate the levels of relevance: highest, moderate, and minimal, respectively.

.

(A) Moon et al. analyzed temporary housing on Yeonpyeong Island, identifying economic feasibility, health standards, structural safety, and site conditions as primary considerations [35]. They emphasized cost-effective design and rapid assembly as critical during early disaster response, and argued that basic health infrastructure and stable structural elements are essential during emergencies.

(B) Wang et al. compared temporary housing in Korea and Japan, highlighting livability, mobility, responsiveness, and sustainability as key factors [36]. The authors discussed the importance of securing communal spaces, pre-identifying potential sites, and developing site plans. They proposed improvements in overall site planning and long-term maintenance.

(C) Moon and Lee analyzed diverse types of temporary housing through domestic and international case studies, focusing on environmental, social, and technological characteristics [37]. Their study revealed that different disaster types require different housing approaches, and that long-term stability can be achieved through socially connected, culturally appropriate designs. It also presents application examples and limitations of various models, stressing the need for efficient resource use and adaptive design strategies.

(D) Kim and Nam examined natural disaster cases in Asia, focusing on structural safety and cultural considerations [38]. They emphasized how spatial design can support social integration and psychological recovery following a disaster, and advocated modular structures and region-specific designs. Their study further emphasized the importance of spatial flexibility and durability in enabling an efficient and economical disaster response.

(E) Stocker et al. analyzed 66 international cases, emphasizing long-term material selection, environmental compatibility, and sustainability [39]. They proposed increasing adaptability and the use of recyclable materials to reduce waste and stressed the need for designs that align with local environmental demands.

These studies provide a multifaceted analysis of the environmental factors necessary for the design of temporary housing and demonstrate how design objectives interact with disaster-specific characteristics. By integrating these factors, it is possible to propose actionable design directions that are adaptable to various disaster scenarios.

3.2. Derivation of Prompt Items for Generating Temporary Housing Exterior Design Alternatives

Based on the analysis of environmental factors in Section 3.1, a comprehensive set of potential design considerations was identified. However, to create a focused and effective methodology for a Text-to-Image (T2I) workflow, it was necessary to distill these factors into a core set of prompt items. The primary selection criterion was direct visual representability—whether a factor could be translated into concrete visual attributes in a generated image of a building’s exterior.

Following this criterion, five key prompt items were selected, as shown in

Table 2. Prompt items for generating exterior design alternatives for temporary residential facilities.

Label	Prompt Item
A	Type of Disaster
B	Location and Climate
C	Duration of Residence
D	Materials and Structure
E	Housing Design

: Type of Disaster, Location and Climate, Duration of Residence, Materials and Structure, and Housing Design. These elements were chosen because they fundamentally and directly influence the visual form, material texture, and environmental context of a temporary structure. Conversely, other important factors identified in Table 1—such as Economic Feasibility, Hygiene and Health, and Social Sustainability—were intentionally excluded. While critical for the overall success of a housing project, these factors are abstract, non-visual concepts that a T2I model cannot effectively render in an architectural image. This deliberate selection focuses the study on its core objective: rapidly generating tangible, visual design alternatives for initial conceptual exploration.

Table 2 presents a basic framework of prompts that incorporates each of these elements, enabling the generation of T2I-based temporary housing design alternatives. Based on this framework, detailed prompts can be developed for each item in the event of an actual natural disaster to guide image generation.

Type of Disaster: This category includes a range of natural hazards, such as earthquakes, floods, typhoons, strong winds, landslides, volcanic eruptions, tsunamis, and desertification. Each disaster type presents distinct challenges that inform the selection of appropriate structural systems and materials.

Location and Climate: Geographical and climatic conditions significantly influence exterior design alternatives. For instance, hot and humid climates require well-ventilated and shaded structures, while cold environments demand thermal insulation. Terrain characteristics—such as coastal, mountainous, or arid regions—also impact material and structural decisions, ensuring designs are environmentally appropriate and functionally efficient.

Duration of Residence: Occupancy duration is categorized as short-term (up to 1 month), medium-term (3 months to under 1 year), or long-term (1–3 years). The expected length of stay influences design decisions regarding durability, material selection, and cost efficiency.

Materials and Structure: Options include wood, stone, metal, adobe/mudbrick, bamboo, recycled plastic, cardboard, and containers, alongside structural systems such as lightweight steel frames, timber frames, masonry, and tensile membrane structures. Material choices are evaluated based on durability, local availability, and environmental sustainability.

3.3. Development of Hypothetical Scenarios for Generating Temporary Housing Exterior Design Alternatives

To ensure the practical relevance of the generated design alternatives, this study developed five hypothetical disaster scenarios grounded in a scientific basis. The selection of these scenarios was directly informed by authoritative global disaster data. According to the Centre for Research on the Epidemiology of Disasters (CRED) in its 2023 Disasters in Numbers report [3] and the United Nations Office for Disaster Risk Reduction (UNDRR) in its Human Cost of Disasters report [4], floods, storms, earthquakes, droughts, and landslides consistently rank among the most frequent and impactful natural disasters worldwide in terms of human cost and economic damage. These five disaster types were therefore chosen as the foundation for our scenario development. Each of the five scenarios (see

Table 3. Hypothetical scenarios and prompts for temporary residential facilities.

No.	Disaster Situation	Hypothetical Scenario		Prompt
1	Flood	A	Severe flood conditions, rapid water level rise	Modular container-based temporary shelter, made from waterproof and insulated materials, elevated on steel stilts, designed for flood-prone areas with rapid water level rise, tropical climate, high rainfall, strong steel frame, sustainable design, minimalistic interior, durable for medium-term residence, rapid assembly, disaster-relief housing High quality, photorealistic, detailed design, sturdy, weather-resistant, clean lines, modern aesthetic, innovative structural elements Blurry, poorly constructed, low quality, unstable, weak materials, traditional tents, outdated design, fragile structure
		B	Low-lying coastal area, tropical climate, high rainfall
		C	Medium-term
		D	Recycled steel and plastic materials for durability and waterproofing
		E	Elevated modular container homes designed for flood-prone areas
2	Storm	A	Strong winds and cyclone-prone conditions	Aerodynamic, wind-resistant modular shelter, designed for cyclone-prone areas, reinforced steel frames, lightweight aluminum panels, waterproof construction, coastal climate with high humidity and frequent storms, designed for short-term residence, easy deployment, sustainable materials, compact design High quality, photorealistic, sharp details, robust structure, clean lines, modern design, windproof, storm-resistant Blurry, poorly designed, low quality, unstable, weak joints, fragile materials, poorly assembled, generic structure
		B	Coastal region with high humidity and frequent storms
		C	Short-term
		D	Reinforced lightweight materials to resist wind pressure
		E	Aerodynamic dome-shaped housing for storm resilience
3	Earthquake	A	High-magnitude earthquake with significant structural damage	Single-story modular housing, earthquake-resistant design with shock-absorbing joints, reinforced concrete and steel materials, designed for high-magnitude earthquake-prone regions, urban area, medium-term housing, flexible connectors for seismic activity, durable construction, sustainable and secure design High quality, photorealistic, sharp, detailed design, strong structural integrity, modern appearance, innovative engineering Blurry, poorly constructed, fragile, unstable, low-quality materials, weak connectors, generic and outdated designs
		B	Seismically active urban area with high population density
		C	Medium-term
		D	Earthquake-resistant materials such as reinforced concrete
		E	Modular units with flexible joints to withstand seismic activity
4	Drought	A	Extreme heat and water scarcity in arid regions	Dome-shaped modular shelter, highly insulated, designed for extreme heat and arid desert climates, made with bamboo and mud-based materials, equipped with rainwater harvesting systems, natural ventilation for sustainable cooling, long-term residence, eco-friendly design, solar-powered for energy efficiency High quality, photorealistic, sharp details, eco-friendly, innovative, sustainable, clean design, vibrant and modern Blurry, poorly constructed, unstable, low-quality, damaged materials, weak insulation, generic structure, ineffective cooling
		B	Desertified climate with prolonged drought conditions
		C	Long-term
		D	Bamboo and mud-based eco-friendly materials
		E	Insulated dome-shaped housing with integrated cooling systems
5	Landslide	A	Heavy rainfall causing unstable slopes and mudslides	Portable modular shelter, reinforced for slope stability in landslide-prone areas, made with lightweight and durable aluminum, waterproof fabric materials, integrated drainage systems, elevated design for heavy rainfall regions, compact and rapid deployment for short-term emergency response High quality, photorealistic, sharp, detailed design, sturdy and modern, innovative, weather-resistant, clean structure Blurry, poorly designed, fragile, unstable, low-quality, weak drainage systems, generic and outdated construction
		B	Mountainous region with high precipitation
		C	Short-term
		D	Lightweight aluminum and waterproof fabric materials
		E	Elevated housing units on reinforced beams to prevent landslide damage

Note: A, B, C, D, and E represent specific categories of relevance: type of disaster, location and climate, duration of residence, materials and structure, and housing design.

) was developed using the five prompt elements defined earlier—type of disaster, location and climate, duration of residence, materials and structure, and housing design—to guide the generation of design alternatives using T2I technology. The detailed content of each scenario was formulated using ChatGPT-4, incorporating specific environmental conditions and design elements to ensure that the resulting designs were appropriate for disaster situations.

To illustrate, Scenario 1 (Flood) is described in further detail as follows: Flooding is one of the most frequently occurring natural disasters globally, often caused by heavy rainfall and river overflow, particularly in tropical and coastal regions [3]. Reflecting these conditions, the scenario was set in a low-lying tropical coastal area with a high flood risk. The intended duration of residence was medium-term. The recommended materials and structural features included waterproof and thermally insulated recycled shipping containers supported by steel frames. The proposed housing design consisted of modular units elevated on stilts to mitigate potential flood damage.

4. Deriving Temporary Housing Exterior Design Alternatives Using T2I Technology

4.1. Workflow Development for Generating Temporary Housing Exterior Designs with T2I Technology

To generate exterior design alternatives for temporary housing based on the hypothetical scenario prompts defined earlier (Table 3), a workflow was constructed using Stable Diffusion ComfyUI, as illustrated in Figure 2. For the core generative model, Stable Diffusion v1.5 [40] was selected over more recent versions such as SDXL. While newer models may offer superior native resolution and prompt adherence, the primary objective of this study was to develop a systematic and reproducible . For this purpose, the v1.5 model was deemed more suitable due to its mature and extensive ecosystem of supporting tools (e.g., ControlNet, custom nodes) that allow for precise, fine-grained control over the generation process. This robust compatibility is crucial for establishing a replicable workflow, which was prioritized over the raw image fidelity of newer, less-established models.

Figure 2 presents the workflow configuration used in this study within the Stable Diffusion ComfyUI interface. The workflow is divided into four main modules—Model, Diffusion, Upscale, and Output—each representing a distinct stage of the process. These modules are further detailed in the following figures to illustrate their specific setups. The use of Set and Get nodes streamlined parameter handling, ensuring greater clarity and control within the workflow. The workflow automates the multistep process of generating exterior designs for temporary housing. Each stage requires specific input values, and the final output consists of design images adjusted according to user-defined parameters.

The first step in the ComfyUI workflow (Figure 3) utilized in this study is the Model Setup process, which lays the groundwork for image generation. This process begins with the Load Checkpoint node to load the image generation model. Subsequently, the CLIP Text Encoder node is used to convert the prompts, derived from Table 3 of this paper, into embedding vectors that the AI model can interpret. At this stage, a positive prompt containing key design elements such as the form, material, and texture of the temporary structure, and a negative prompt to exclude undesirable image qualities, are entered. Notably, the positive prompt was divided into two separate CLIP Text Encoder nodes: one for the architectural design elements and another for the overall quality and detail of the image. This was done to encourage the AI model to interpret and incorporate the content of each prompt more independently. These two positive embedding vectors are then combined into a single, unified conditional data set via the Conditioning Concat node. The negative prompt, which only includes quality-related conditions, was processed using a single node. The model, CLIP, VAE, and the generated positive and negative embedding data are then consolidated into a single pipe via the Set node and passed on to the subsequent image generation stage. This module plays a crucial role in moving beyond simple text-to-image generation and precisely controlling architectural features to produce results that align with the research objectives.

The second stage of the workflow (Figure 4), the Diffusion Module, performs the core image generation process by utilizing the pipeline information saved from the previous module. This module uses a Get node to retrieve the stored data and then employs an Empty Latent Image node to create a noise tensor that serves as the initial canvas for image generation. This node determines the pixel resolution of the latent space, which is then connected to the subsequent KSampler node.

The KSampler node is where the central operation of the diffusion model takes place. It progressively denoises the latent space to generate an image based on the given conditions. To control the image quality and style, several parameters were carefully configured. The number of sampling steps, which adjusts the iteration count for image creation, was set between 30 and 40. The Classifier-Free Guidance (CFG) value, which controls how strongly the model adheres to the prompts, was set between 7.0 and 8.5 to ensure the prompts were faithfully reflected. The sampler was set to euler and the scheduler to karras to efficiently manage the diffusion process. Finally, the denoise value was set to 1.0 to maximize the intensity of the noise reduction, thereby enhancing the final image quality.

Ultimately, a VAE Decode node restores the generated latent representation into a pixel-based image, outputting the final result. This outcome is then stored using a Set node, making it available for the subsequent upscaling module. This module, through its precise parameter settings, successfully generates high-quality exterior images of temporary structures.

The third stage of the study, the Upscale Module (Figure 5), functions to enhance the resolution of the image generated in the preceding diffusion process. To improve the resolution of the retrieved image, the Upscale Model Loader node is used to load an upscaling model. The workflow then uses the Image Upscale with Model Batched node to process the image in batches, performing the upscaling. This workflow specifically utilized the 4x-UltraSharp.pth model to upscale images in 16-unit batches. The Upscale Image By node then rescales the 4x-upscaled images to their original size, producing the final output. This module is essential for maximizing the visual quality of the images and, thus, the overall completeness of the research results.

The final stage of the study, the Output Module (Figure 6), serves to store and analyze the final results. This module utilizes a Get node to retrieve the image generated in the preceding process. The finalized output is then saved to a specified directory via the Save Image node, thereby completing the entire automated image generation workflow. This module plays a vital role in evaluating and preserving the ultimate value of the research outcomes.

4.2. Generation of Temporary Housing Exterior Design Alternatives

Table 4. Generated temporary design alternatives through the workflow.

Scenario	Alternative 1	Alternative 2	Alternative 3
Flood
Storm
Earthquake
Drought
Landslide

Note: Alternatives 1, 2, and 3 correspond to small-scale housing for one to two occupants, medium-scale housing for four to five occupants, and large-scale housing for eight to ten occupants, respectively.

summarizes the exterior design alternatives generated for each of the five scenarios using the workflow outlined in Section 4.1 (Figure 2). For every scenario described in Table 3, three design alternatives were produced, corresponding to different occupancy sizes: Alternative 1 (small-scale housing for one to two occupants), Alternative 2 (medium-scale housing for four to five occupants), and Alternative 3 (large-scale housing for eight to ten occupants). By differentiating structural and spatial attributes according to occupancy size within the same disaster context, the study assessed whether GenAI could effectively reflect such distinctions in the generated design images. The objective was not only to explore visual diversity but also to examine whether the designs aligned with practical requirements and end-user needs.

The prompt composition remained consistent across all alternatives within each scenario; however, occupancy-specific keywords were varied to reflect scale. For small-scale housing (1–2 occupants), keywords such as “compact size” and “small scale” were used; for medium-scale housing (4–5 occupants), “medium size” and “medium scale” were applied; and for large-scale housing (8–10 occupants), “large size” and “huge scale” were included. Due to the inherent limitations of Stable Diffusion, not all keywords were consistently reflected in the generated images. To achieve the desired scale and form, prompt keywords were modified, the weights of specific terms were adjusted, and the image generation process was repeated as necessary.

Small- and large-scale housing units exhibited clear differences in size and were generally generated successfully. However, medium-scale units occasionally resembled either small or large configurations. Nonetheless, by refining prompt keywords, adjusting weights and repeating the generation process, occupancy-specific scale distinctions were effectively captured in the resulting images.

Beyond scale differentiation, each scenario reflected relevant material and environmental considerations. For example, the flood scenario featured waterproof, moisture-resistant recycled shipping containers with steel frames. The storm scenario utilized wind-resistant lightweight metal panels. Earthquake-resilient designs integrated reinforced concrete and structural steel for seismic resistance. Drought-related designs employed sustainable materials such as bamboo and adobe bricks, while the landslide scenario utilized lightweight aluminum and waterproof fabric tents for quick deployment and adaptability.

These results suggest that GenAI-based design tools can effectively accommodate the diverse design requirements across disaster types and occupancy levels. The material selections and structural features tailored to each scenario highlight the potential applicability of this method to real-world emergency housing contexts. Compared to conventional methods, this approach enables significantly faster and more efficient generation of design alternatives. Additionally, GenAI facilitates rapid customization in the early design stage by considering multiple variables (e.g., disaster type, occupancy, and environmental conditions), thereby enhancing its practical utility in time-sensitive disaster-response efforts. Future research should focus on validating the structural stability and real-world applicability of AI-generated alternatives and expanding the scope to include context-aware design capabilities through image-to-image (I2I) techniques for more resilient architectural solutions.

5. Discussion

This section critically examines the practical applicability of the design alternatives proposed by Generative AI. Section 5.1 presents empirical data from a survey of architectural experts, outlining the strengths and weaknesses of the AI-generated images (the ‘what’). Subsequently, Section 5.2 provides an in-depth analysis of the underlying reasons for these findings (the ‘why’), focusing on the ‘feasibility gap’ between aesthetic proposals and buildable solutions from the perspective of material and engineering limitations.

5.1. Survey-Based Evaluation of Generated Design Alternatives

To empirically validate the practical applicability of the design alternatives generated by Generative AI (GenAI), this section details an online survey conducted with architectural design experts. The primary objective of this survey was to evaluate the AI-generated temporary housing designs from multiple perspectives and to derive insights for future improvements and real-world applications.

The survey was designed to assess the generated images based on three core criteria: (1) Prompt Fidelity, (2) Architectural Plausibility, and (3) Practical Feasibility.

For this evaluation, we first selected the target respondents. A group of 30 experts with over 10 years of experience in architectural design, temporary housing, and modular construction was established. These professionals were either currently involved in or had prior experience with construction projects. The survey was created using Google Forms, and the 30 selected experts were contacted in advance. A link to the Google Form was then distributed via email, inviting them to participate. The survey was conducted online for approximately two weeks, from September 4 to September 12, 2025, during which all 30 recipients responded, yielding a valid sample for analysis.

The survey used in this study consisted of two sections: a quantitative evaluation and qualitative feedback. In the quantitative section, respondents were asked to evaluate 15 images based on the three core criteria. The qualitative feedback section included open-ended questions to gather expert opinions on the potential applications, limitations, and areas for improvement of GenAI in the field of disaster response architecture. As shown in

Table 5. Overall Evaluation Scores for Generated Designs.

Criteria	Number of Images (N)	Minimum (Min)	Maximum (Max)	Mean (M)	Standard Deviation (SD)
Prompt Fidelity	15	4.2	6.8	5.80	0.95
Architectural Plausibility	15	2.5	5.9	4.50	1.32
Practical Feasibility	15	1.8	5.2	3.90	1.48

, the analysis of the survey responses revealed distinct strengths and weaknesses of the GenAI-based design process.

(1) Prompt Fidelity: This criterion measured how well the generated images reflected the specific keywords in the prompt, such as disaster type, materials, and structural concepts. The designs generally received high scores in this area (mean of 5.8/7.0). For example, the ‘flood’ scenario successfully generated images of container structures on stilts, while the ‘drought’ scenario produced dome-shaped shelters made from materials like bamboo and mud. This indicates that the text-to-image (T2I) model effectively translated key textual descriptors into corresponding visual elements. However, some inconsistencies were found in reflecting scale variations—such as small, medium, and large—which required manual adjustments and iterative generation to achieve the desired output.

(2) Architectural Plausibility: This item assessed whether the designs were architecturally coherent and rational. The average score was moderate (4.5/7.0). Experts noted that while the designs were visually appealing and stylistically appropriate for initial concepts, they often lacked the structural integrity or detailed connections essential for real buildings. For instance, the earthquake-resistant design showed a robust form made of reinforced concrete, but the specific appearance of the shock-absorbing joints mentioned in the prompt was not clearly visualized. This suggests that while GenAI excels at creating conceptual aesthetics, it currently has limitations in generating technically detailed and structurally sound architectural solutions without further refinement.

(3) Practical Feasibility: This criterion evaluated the real-world applicability of the designs, considering factors such as ease of construction, material availability, and cost-effectiveness in a disaster situation. This category received the most critical feedback (mean of 3.9/7.0). Experts pointed out that while installing temporary housing on water in a flood zone was an interesting idea, it was unrealistic, as flood victims would prefer to reside on safe, solid ground. For the ‘storm’ scenario, some noted that features like long, overhanging steel roofs appeared vulnerable to high winds. Although the ‘landslide’ scenario designs appropriately used lightweight materials like aluminum and waterproof fabric for rapid deployment, experts pointed out that the complexity of some generated forms could hinder the speed of assembly required in an emergency. In summary, the qualitative feedback consistently emphasized that while GenAI is a powerful tool for rapid ideation, its outputs must be critically reviewed and adapted by human designers to meet the strict functional and logistical demands of temporary disaster housing. The detailed feasibility scores across different disaster scenarios are presented in

Table 6. Feasibility Score Analysis by Disaster Scenario.

Disaster Scenario	Number of Images (n)	Mean (M)	Standard Deviation (SD)	F Value	p-Value (p)
Flood	3	4.85	1.15	4.21	0.42
Storm	3	4.10	1.28
Earthquake	3	3.25	1.42
Drought	3	4.30	1.33
Landslide	3	2.98	1.51
Total	15	3.90	1.48

.

In summary, this survey confirms that GenAI has significant potential to accelerate the initial conceptual design phase for disaster relief architecture. GenAI excels at rapidly generating a diverse range of scenario-specific design alternatives that align well with initial text-based prompts.

However, for these designs to be truly viable, a considerable level of human expertise is required to bridge the gap between AI-generated concepts and buildable solutions. The findings of this study suggest that the current role of GenAI is best defined as a ‘collaborative partner’ that enhances the creativity and efficiency of architectural professionals, rather than a replacement for their critical thinking and technical knowledge. This view is consistent with expert commentary that AI lacks the cultural understanding, empathy, and ethical judgment required to replace architects, reinforcing its role as an augmentative collaborator rather than a substitute [41].

Future research should focus on developing workflows that integrate engineering constraints, building codes, and user-centered functional requirements directly into the generation process to enhance the architectural plausibility and practical feasibility of the outputs.

5.2. Analysis of Durability and Service Life in AI-Generated Temporary Housing

While temporary housing is intended for short-term use, it often remains occupied for extended periods. Therefore, material durability and service life are critical issues directly linked to the residence’s resilience. This section critically examines the extent to which the design alternatives visualized by GenAI reflect these practical constraints and identifies their inherent vulnerabilities. The generated images effectively visualize each scenario’s core concept; however, they exhibit a clear limitation in overlooking the essential engineering details required to ensure long-term material durability.

A primary issue is the superficial representation of materials. In the flood scenario, for instance, the AI consistently proposed elevated structures to prevent inundation damage. However, it failed to represent the primary vulnerability of steel containers: corrosion. The containers in the images have unrealistically smooth surfaces, whereas, in humid climates, engineering measures such as anti-corrosion coatings, sealed finishes on welded joints, and effective drainage design are essential to prevent rust. The AI implemented the macro-level concept of ‘avoiding floodwaters’ but missed the practical challenge of ‘protecting materials from moisture.’ Similarly, the designs for the drought scenario, which used bamboo and mud-brick, depicted these natural materials in an overly idealized manner. In reality, without protective elements like deep eaves or solid foundations to shield them from rain and ground moisture, these materials can decay or erode within a few years. The AI’s outputs emphasized the aesthetic aspects of the materials while neglecting the protective measures essential for long-term use. For the storm and landslide scenarios, the proposed alternatives using lightweight aluminum and waterproof fabric are advantageous for rapid installation. However, the images do not show the robust foundations, anchoring, and joints necessary for structural stability. Lighter structures are more susceptible to high winds and ground shifts, making their anchoring methods critical, yet the AI only presented a plausible overall form while omitting the structural details directly related to safety.

These findings suggest that current T2I technology is a powerful tool for rapidly visualizing diverse ideas in the early design stages, but it does not yet reflect the physical limitations of materials or the engineering details that ensure long-term durability. This means that AI-generated designs currently remain at the level of ‘conceptual ideas.’ Consequently, it is crucial to recognize the gap between these digital concepts and their physical implementation. Material suggestions, such as “bamboo and adobe” for the drought scenario, are generated based on stylistic and semantic associations within the training data, not on a rigorous analysis of local resource availability, supply chains, or construction costs. Therefore, these AI-generated outputs should be considered a starting point for the design process, requiring strict review by architects, engineers, and local stakeholders to ensure their feasibility, cost-effectiveness, and appropriateness within a specific post-disaster context. The role of the human expert as a validator and contextualizer in this workflow is essential.

Furthermore, this study acknowledges the potential for inherent biases within GenAI models. Recent empirical analyses show that large web-scraped image datasets are English- and Western-centric, which in turn biases text-to-image outputs toward Western cultural representations [42]; complementary evidence of Western bias has also been reported for large vision–language models [43]. The training data for large-scale models like Stable Diffusion is predominantly sourced from the internet, which may lead to an over-representation of Western architectural styles. This could result in designs that are culturally or climatically inappropriate for certain non-Western regions. Mitigation strategies include careful prompt engineering with region-specific cultural cues (e.g., “indigenous Filipino bamboo house”). However, a more robust long-term solution, representing a direction for future research, involves fine-tuning models on curated datasets of regional architecture using techniques like Low-Rank Adaptation (LoRA) to generate more culturally and contextually sensitive designs.

To overcome these limitations, future research must progress toward enhancing the technical specificity of prompts. This involves moving beyond simply naming materials to actively incorporating specific engineering requirements that improve durability and structural stability, such as “corrosion-resistant coating,” “UV-protective finish,” and “reinforced foundation joints.” Through such prompt refinement, GenAI can evolve from a tool for creating aesthetically pleasing images into a practical instrument for proposing genuinely buildable and sustainable architectural alternatives.

6. Conclusions

This study investigated the application of Generative AI for rapidly generating exterior design alternatives for temporary housing. The findings demonstrate a critical duality: while the T2I workflow successfully produced a diverse range of creative, scenario-specific concepts, expert evaluations revealed a significant ‘feasibility gap’ between these visual ideas and structurally sound, buildable solutions. The core contribution of this research, therefore, is not merely the demonstration of a novel workflow, but the empirical identification and analysis of this gap, which is crucial for understanding the current role of GenAI in disaster-response architecture.

This study offers meaningful insight into the applicability of GenAI as a viable tool in architectural design, particularly in its capacity to produce customized design alternatives rapidly during the early stages of planning. By accommodating scenario-specific variables, this method contributes to the enhancement of disaster response systems and lays the groundwork for innovative developments in architectural design.

Nonetheless, the study has certain limitations. The design outputs generated by GenAI may not fully reflect the functional, ergonomic, or daily living requirements of displaced populations. Moreover, the unequal weighting of input prompts in Stable Diffusion affects the consistency and applicability of renderings in real-world scenarios.

Despite these limitations, this study serves as an important starting point that demonstrates the real-world applicability of AI-based design methodologies and lays the groundwork for future advancements.

Future research should focus on incorporating design constraints informed by actual building codes, regional environmental factors, and user needs. Practical design alternatives must also consider both functional viability and environmental performance. Additionally, the application of I2I technology should be explored to develop more realistic design outputs that align more closely with the installation environment. Finally, in-depth studies are needed to integrate user-centered design elements and architectural performance requirements.

In sum, this study successfully establishes a point of convergence between generative AI and architectural design practice. It confirms that GenAI’s greatest value is not as an autonomous agent, but as a ‘collaborative partner’ that amplifies the creativity of human designers. However, bridging the gap between AI-generated concepts and real-world application requires the indispensable expertise of architectural professionals. Future development must therefore focus on integrating engineering constraints to foster a more seamless synergy between human and artificial intelligence in creating resilient architectural solutions.