46 | History, Theory & Criticism CTBUH Journal | 2023 Issue III
Introduction
Architects and urbanists have long been
fascinated by the potential for new
technologies to shape the built
environment—the emergence of Artificial
Intelligence (AI) is no exception. Some would
argue, like in many other industries, AI has
the potential to drastically alter architecture,
urbanism, and the future of cities. Others,
however, view the new technology as only
potentially transformative and less of an
existential threat. As discourse over AI
technology continues, it is imperative that
experts explore the technology and consider
its true impact. Foster + Partners, the largest
architecture firm in the United Kingdom, has
been exploring AI and machine learning
tools in the early conceptual design phase of
its projects. “In addition to using reference
images or studying precedents online, we
can now use these tools to quickly illustrate
an idea or a feeling that we want a particular
space to evoke,” says Martha Tsigkari, head
of Applied R+D at Foster + Partners. “This
side of machine learning is shaping up to
be a useful tool for drawing inspiration”
(Berg 2022).
Architectural Design and T2I AI
Like many creative endeavors, architectural
design is not about flashes of inspiration or
sudden insights, but rather the result of
persistent effort, experimentation, and
refinement. Architects and designers embrace
the complexity of the world around them and
recognize that the built environment is not a
passive backdrop, but an active agent that
shapes and is shaped by our actions. Style and
form are not sufficient to address the complex
challenges we face today—from technical
and economic to social and environmental—
as they require rigorous analysis, creative
thinking, and disciplined execution.
In a world where the challenges we face are
increasingly interconnected and urgent, it is
important for architecture and design to be
more than an expression of aesthetics, but a
vital tool for confronting issues such as
climate change and social inequality.
Ultimately, it is our ability to ask the right
questions, uncover the needs of our clients
and communities, and respond with
intelligent, thoughtful solutions that set true
design apart from Artificial Intelligence (AI).
Text-to-image (T2I) AI generates visual
Fantasy on Demand: The Temptation
Of Text-to-Image AI
History, Theory & Criticism
Author
Julia Nagele, Senior Principal and Director of Design
Hewitt Architects
101 Stewart St. #200
Seattle, WA 98101, United States
t: +1 206 624 8154
hewittseattle.com
Julia Nagele brings more than 20 years of
experience to her role as senior principal and
director of design at HEWITT Architecture. Her
designapproach in HEWITT Architecture’s classic
studio environment is similar to the project-
based learning she leads as an aliate assistant
professoratthe University of Washington’s College of
Built Environments.
Known for her design of tall buildings, Nagele’s
40-story Emerald Tower, completed in 2020, is
the tallest building on the West Coast created by
an all-female design team. Mama Tower, which is
moving through permitting, will reach 484 feet
(147.6 meters). Nagele graduated Cum Laude with
a Bachelor of Arts in Architecture from Lehigh
University and completed her Master of Architecture
degree from the University of Maryland’s School of
Architecture, Planning and Preservation.
Julia Nagele
Abstract
This paper explores the implications of text-to-image artificial intelligence (T2I
AI) on architectural design practice. It begins with an overview of generative
artificial design. It then explains of why T2I AI is the focus of this study, and the
potential impact on design methodology and aspects of day-to-day practice of
architects and designers. We draw parallels between the effects of T2I AI and
past architectural and urban design futurists to situate this technology within a
larger historical context. Finally, we test the effects of T2I AI on architectural
design methodology, demonstrating its potential for architects to conceptualize,
design, and communicate ideas. Architects should actively explore the
possibilities of T2I AI as a design tool but understand its limitations. T2I AI can
assist designers and stakeholders through visual inspiration, but it is not a tool to
replace the human expertise required to identify, address, and solve complex
challenges in the built environment.
Keywords: Architectural Design, Artificial Intelligence, Text-to-Image
“
The ability to use
T2I AI as a tool to
spark new and
innovative ideas is
like having an extra
team member with a
unique perspective.
”
History, Theory & Criticism | 47CTBUH Journal | 2023 Issue III
content by using a combination of
algorithm-learning machines, neural
networks, and image processing
techniques, but has no understanding of
the norms and values of serious
architectural design methodologies. It has
no awareness of the broad range of
complex issues architects, designers, and
urban planners need to address.
The basic process of creating an image in T2I
AI starts with a text prompt on a Discord
server. In a message prompt, a user begins
by typing “/image,” which activates popular
AI platform Midjourney’s image-generating
bot. One could enter only a few words or a
lengthy description, but a common format is
to begin by defining the desired image’s
style, the subject, adjectives or modifiers, and
ending with specific parameters. Parameters
allow users to shape the image options, add
more importance or”weight”to some
descriptive words than others, or set a
specific aspect ratio (Wolff 2023). After
inputting the description, four thumbnails
are generated (see Figure 1). From there, the
user can either “upscale” one or more of the
images to generate larger, more detailed
versions of them (see Figure 2) or prompt the
bot to provide four more varieties of any of
Figure 1. An initial set of thumbnails created on the rst round of data input into
Midjourney.
Figure 2. An upscaled, more detailed iteration of the initial Midjourney rendering.
the four original images. This offers a quick
and easy way to generate multiple options
and explore different design possibilities. The
process can be repeated as many times as
needed until the desired image and
description are obtained.
T2I AI can create high-quality images with
specific detail in a broad range of styles and
character. The images can be anything, from
a landscape to an architectural structure, or
even a piece of furniture. Viewpoints can be
from any perspective or angle. The images
are then created through a neural network
that has been trained using a vast dataset of
images, allowing it to understand the
patterns and features of nearly anything. The
bot functions as a facilitator of collaboration,
using all the Internet’s graphic content to
stitch together pixels to form a collage.
Because of the ease, speed, and ability of the
bot to generate images, it is an enticing tool
for testing the breadth of a limitless
architectural future, rather than replicating
present conventions.
There are several nuances in the process of
creating the images. First, the selection of
keywords and phrases has a significant impact
on the outcome of the generated images.
Using conventional descriptions such as
“building,” “city,” and “street” led to more
predictable, mundane, and in some cases
incoherent expressions of the built
environment.Attempts to create derivatives
or replicate existing concepts were also
challenging. This point was illustrated after
entering a description of the real-world,
futurist city proposal, The Line, which is
planned in the Kingdom of Saudi Arabia. The
Line is a linear megacity wrapped in a
mirror-like material that is projected to hold
nine million people (see Figure 3). As a purely
graphic exercise, its overall features are clear
and distinctive. It is 500 meters high, 200
meters wide, and 170 kilometers long.
Attempting to generate a large, mirror-
walled structure with similar proportions as
The Line using T2I AI was challenging, as the
scale of the generated image was often
underwhelming (Figure 4), but if “mirror” was
substituted with other words, the results
displayed nondescript coastal cityscapes
(Figure 5). More promising attempts were
created using words and phrases that were
either more abstract or not typically
associated with architecture. Using words
such as “hot-air balloon,” “greenery,” or
“interstellar,” led to a broader interpretation
48 | History, Theory & Criticism CTBUH Journal | 2023 Issue III
Figure 4. Using keywords describing The Line literally, the AI-generated image bore
little resemblance to the original, particularly in terms of scale.
Figure 5. Removing the word “mirror” from the text developed non-descript coastal
cityscapes.
and expression of believable concepts, even
if they were rooted in fantasy.
Another useful tactic was to input well-
known architectural and urban design
sources, precedents, and/or architects to the
image generator. This resulted in a clearer
outcome that seemed to bridge science-
fiction imagery with futuristic architecture.
Providing examples of notable architects and
their work was useful for Midjourney as an
information source, similar to asking ChatGPT
to write something in the style of a particular
author. This is underscored by the work of
Zaha Hadid Architects, where designers are
encouraged to use T2I AI and enter “Zaha
Hadid” as part of their working process,
which creates a feedback loop of design
exploration (Budds 2023).
Human decision and interaction are still
necessary through this whole process;
curation and discernment of the images;
how they are refined, which are selected, and
what purpose they might ultimately serve
are all factors that cannot be decided by
artificial intelligence. T2I AI is clearly about
image production and not about providing a
framework to create architecture in the real
world. Using T2I AI as a visual tool can be
exhilarating and unpredictable; just a few
keywords can produce a vast array of
outcomes in seconds, each having the
potential to inspire new iterations and ideas.
While T2I AI can provide a visual “sugar rush,”
the emergence of the technology has raised
concerns among architects and educators
about its impact on students and
professionals alike. Educational and
professional architectural practice emphasize
the importance of understanding and
responding to a wide range of real-world
issues. Solely focusing on subjective images
and aesthetics, such as what T2I AI provides,
is not an acceptable approach to addressing
serious needs and issues. Therefore, the
question arises: Does this technology have
any true value for architects, urban planners,
and designers? By slightly altering key
phrases, like pulling a handle on a slot
machine, the possibilities of graphic
Figure 3. Rendering of The Line, the 170-kilometer-long city under construction in Saudi Arabia. © NEOM
History, Theory & Criticism | 49CTBUH Journal | 2023 Issue III
Figure 6. Étienne-Louis Boullée’s Cenotaph for Sir Isaac Newton, 1784, is in some ways
a pre-digital work of AI—its purpose was to express Newtonian ideas and symbolism,
not to be a real building. Source: Miller, 2014
outcomes are extensive, and thus reveal T2I
AI’s greatest power: providing answers to
“what if?” in compelling ways.
T2I AI as Architectural “What Ifs?”
While seductive and instantaneous images
are not the goal of architecture, the value of
T2I AI lies in its ability to inspire new ideas. It’s
a tool free from constraints, which can
facilitate unconventional design solutions.
Proposing grand, audacious, and
intentionally conceptual ideas is not new in
architecture. Throughout history, there has
been a consistent parallel form of expression
and thought, separate from everyday
practice, intended to push the boundaries of
what is possible.
Étienne-Louis Boullée and Claude-Nicolas
Ledoux were French architects who lived
during the 18th century, a time of great
intellectual and artistic innovation in Europe.
Both were leaders of the Neoclassical
movement, which sought to revive the
classical forms and principles of ancient
Greece and Rome. Their designs were bold
and innovative, often characterized by
massive scale and powerful geometry.
One of the most notable works of Boullée
was his Cenotaph for Sir Isaac Newton
(Figure 6), which he designed in 1784, 40
years after Newton’s death (Miller 2014). The
Cenotaph was not intended to be a practical
or functional building, but rather an
avant-garde work of architecture. Boullée’s
design was highly conceptual and was
meant to express the grandeur and
significance of Newton’s achievements
through a monumental, symbolic form.
Ledoux’s most famous project was his 1804
design for the House of Surveyors of the
Loue River (see Figure 7) and was also
intended to be visionary work. The project
was part of Ledoux’s larger plan for a utopian
city called Chaux, which he designed for the
king of Prussia. It was a vision of a self-
sufficient community, with all the necessary
functions of daily life contained within a
single, massive complex. The buildings were
to be surrounded by gardens and farmland,
with a central courtyard containing a
fountain and statue (Etlin 1995).
The work of avant-garde architects such as
Archigram (1961–74) are modern examples
of conceptual thinkers who proposed
several radical and innovative concepts that
challenged notions of architecture and
urban design (Moore 2018). Their ideas
were diverse and futuristic. Some of their
most notable concepts include the “Plug-in
City” to the “Walking City” and the “Instant
City.” The “Plug-in City” was a modular and
adaptable city that could be built and
rebuilt as needed, while the “Walking City”
was made up of giant, self-contained
robots or “pods” that could move around
and reconfigure themselves based on the
needs of their inhabitants. The “Instant City”
(see Figure 8), on the other hand, was a
concept for creating temporary and mobile
structures that could be deployed quickly to
meet the needs of people in emergency
situations. While these concepts may have
seemed far-fetched at the time, they have
had a profound impact on the field of
architecture and continue to inspire new
ways of thinking about the built
environment. By pushing the boundaries of
what is possible in architecture, groups like
Archigram have influenced architects to
explore new possibilities.
Archigram’s influence on real-world
architecture can be seen in many
contemporary buildings and designs that
have been inspired by their futuristic and
unconventional concepts (see TalkingTall,
page 58). For example, one could compare
the design of the Centre Pompidou in Paris
(see Figure 9 )completed in 1977 and
designed by Richard Rogers and Renzo Piano,
to the Plug-in City concept. The 20th-century
high-tech design was driven by the concept
of “inside-out” or “reverse architecture,”
where the building’s mechanical and
functional systems are exposed on the
outside. This design approach allows for
flexible, column-free, and adaptable interior
spaces, with the ability to reconfigure over
time. The building’s colorful pipes and ducts
are not merely decorative but serve a
functional purpose distributing air, water, and
electricity throughout the structure.
Figure 7. Claude-Nicolas Ledoux, House of Surveyors of the Loue River (1804), is the
visionary work of an architect, with no intention to build. Source: Ricci, n.d.
50 | History, Theory & Criticism CTBUH Journal | 2023 Issue III
Figure 9. Centre Pompidou, Paris, designed by Renzo Piano and Richard Rogers, was presaged by Archigram’s
megastructure designs. © Jean-Pierre Dalbéra
Figure 8.
“Instant City” was a concept for creating temporary and mobile structures that could be deployed quickly to
meet the needs of people in emergency situations. © Peter Cook
Another megaproject echoing Archigram’s
out-of-the box conceptual thinking is the
Taipei Performing Arts Center (see Figure 10),
designed by OMA. The multi-functional
cultural center is located in the Xinyi District
of Taipei. The project’s primary concept was
to create a space that seamlessly integrates
the arts and the surrounding urban context.
Archigram’s Instant City concept was a
proposal for a flexible and adaptable urban
infrastructure. Today, large-scale modular
architecture has become a practical solution
for many building projects, offering benefits
such as cost savings, time efficiency, and
sustainability. The modular construction of the
461 Dean (see Figure 11) in Brooklyn, New
York, designed by SHoP Architects and built
by Forest City Ratner, is an example of a
large-scale project that focuses on
prefabrication and adaptability.
T2I AI in Everyday Applications
In architecture, there is value in envisioning
big ideas, and a compelling image generated
by T2I AI can inspire new concepts. AI doesn’t
consider our own self-imposed rules like
personal biases, meaning it could add a fresh
perspective on design problems. The ability to
use T2I AI as a tool to spark new and
innovative ideas is like having an extra team
member with a unique perspective. Several
prominent global architecture firms have
resources to establish in-house research and
development teams to test big ideas and
enter competitions. With the invention of T2I
AI, a readily available, inexpensive tool
becomes accessible for small- and medium-
sized firms to explore.
Designers face constraints, such as client
preferences, budgets, zoning regulations, and
building codes, which, in turn, shape the
design process. T2I AI images cannot consider
real-world constraints and conditions,
allowing for unconventional design ideas and
perhaps even a future real-world solution. AI
technology could assist in creating something
truly innovative.
The ability to rapidly generate images through
AI technology can foster an innovative design
process, involving diverse groups of
stakeholders at an earlier stage. This can
enable architects and designers to better
understand the needs and perspectives of
various groups and incorporate them into
designs that reflect the unique needs of the
community. With the help of AI-generated
images, community members and other
stakeholders can create and share their own
compelling images, facilitating
communication and collaboration throughout
the design process. This ultimately leads to
more informed decisions and a stronger sense
of community investment in the final product.
History, Theory & Criticism | 51CTBUH Journal | 2023 Issue III
Budds, D. (2023). “The Zaha Hadid AI Feedback Loop.”
Curbed. https://www.curbed.com/2023/04/zaha-hadid-
architects-ai-design-midjourney-dall-e-2.html.
Etlin, R. (1995). Symbolic Space. Chicago: University of
Chicago Press.
Hobson, B. “Archigram’s Instant City Concept Enables ‘A
Village to Become A Kind of City for A Week’ Says Peter
Cook.” Dezeen. https://www.dezeen.com/2020/05/13/
archigram-instant-city-peter-cook-video-interview-vdf.
Miller, M. (2014). “AD Classics: Cenotaph for Newton /
Etienne-Louis Boullée.” ArchDaily. https://www.archdaily.
com/544946/ad-classics-cenotaph-for-newton-etienne-
louis-boullee.
Moore, R. (2018). “The World According to Archigram.” The
Guardian, 18 November 2018. https://www.theguardian.
com/artanddesign/2018/nov/18/archigram-60s-architects-
vision-urban-living-the-book.
Ricci, B. (n.d.) “Imagining New Worlds. or How Architecture
Has Dreamt of Utopia.” Artland Magazine. https://magazine.
artland.com/imagining-new-worlds-or-how-architecture-
has-dreamt-of-utopia/.
Wolff, T. (2023). “The Midjourney Cheat Sheet (v5).” Medium.
https://medium.com/design-bootcamp/the-midjourney-
cheat-sheet-v5-54b5fd92d2da.
Conclusion
Despite all the excitement, it’s worth
remembering that AI is still in its infancy and
rapidly evolving. Current versions of T2I AI,
readily available to the public, have limited
usefulness in professional architectural
design. It could aid architects and designers,
but is not a replacement. The most
promising use of T2I AI is early concept and
idea generation or non-specific, open-ended,
conceptual studies. Even with that narrow
application, it should be noted that T2I AI
would not be able to respond to specific
design goals or form concepts around the
particular needs of people, which is
important in the early stages of design. T2I AI
would not be suitable for visualizing
real-world architectural projects with specific
contexts and technical drawings, such as
images needed for successful permitting and
city-level approvals. The risk of project delay,
cost uncertainties, and inaccurate depictions
of projects in high-stakes city approval
processes is too great to use the current
forms of T2I AI applications.
Architectural design and creativity involve
responding to the needs of clients, with a
serious and sustained effort to reveal and
respond to complex social, environmental,
and technical challenges for the human
condition. The strength of T2I AI, and the
visionary imagery it can create, may help
inspire design teams, clients, and city
officials to tackle today’s challenges, but it
cannot solve them. While T2I AI can provide
another tool to add to our design process,
we must be wary of the temptation of
instant, fantastic images of artificial
architecture. However, we should continue
to embrace the unexpected and use it as
one of many starting points for thought,
creativity, and innovation. As designers, we
should be willing to use any tool at our
disposal that could potentially lead to the
betterment of the human condition.
See more examples of Hewitt’s Midjourney
AI explorations at: ctbuh.org/Nagele-
text-2-AI.
Unless otherwise noted, all photography credits
in this paper are to HEWITT Architects.
References
Berg, N. (2022). “AI Tools Like DALL-E 2 and Midjourney
are Helping Architects – and Their Clients – Design
New Buildings.” Fast Company. https://www.
fastcompany.com/90780871/ai-tools-like-dall-e-2and-
midjourney-are-helping-architects-and-their-clients-
design-new-buildings.
Figure 10. The Taipei Performing Arts Center by OMA, is another megaproject echoing Archigram’s out-of-
the-box conceptual thinking. © Supanut Arunoprayote.
Figure 11. The prefabricated 461 Dean project in Brooklyn, New
York City, can be seen as a latter-day advancement of Archigram’s
Instant City. © Kidy182 (cc by-sa)
