INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
334
DESIGN GENERATION BASED ON ARTIFICIAL INTELLIGENCE: A COMPARATIVE
ANALYSIS OF METHODOLOGICAL APPROACHES
Zilola Rakhmatillaeva
Tashkent University of Architecture and Civil Engineering
Abstract:
This article explores four methods of generating architectural and interior design using
artificial intelligence — zero-shot, one-shot, few-shot, and prompt chaining — from both theoretical
and experimental perspectives. Practical experiments were conducted based on real design tasks to
compare output quality, user control, technical complexity, and generation speed. Particular focus is
given to the prompt chaining approach, which was used to develop a step-by-step landscape design
concept for the Shorkul reservoir area in Bukhara. The article outlines the advantages and limitations
of each method and offers recommendations for effective integration of AI tools into the design
process.
Keywords:
artificial intelligence, design generation, prompt chaining, zero-shot, one-shot, few-shot,
architectural design, landscape project, ChatGPT, Midjourney, AI-powered interior design
Introduction:
Artificial Intelligence (AI) is fundamentally transforming design processes. In
traditional design workflows, engineers and architects are often limited to exploring only a few
alternatives due to constraints in time and resources. In contrast, generative AI models are capable of
rapidly producing countless design solutions [1]. Research shows that AI not only enhances technical
optimization in architecture and design, but also improves creative processes — enabling designers to
explore multiple alternative solutions through decision-support and generative systems, all while
maintaining creative control in the hands of the human designer [1].
Using AI tools, designers can quickly generate, evaluate, and revise multiple design alternatives,
which facilitates better-informed decisions at early stages of the design process [1]. Currently, AI-
generated solutions are gaining critical importance in fields such as interior design, architecture, and
landscape design. Designers frequently need to consider a wide range of options in terms of materials,
colors, and forms — a process that can be time-consuming. AI helps accelerate this process and
expands the scope of creative exploration.
For instance, in a 2024 study by Gallega and Sumi, ChatGPT and DALL·E 2 were tested to streamline
the selection of materials and textures in 3D interior scenes [2]. In this experiment, professional
designers rated AI tools 72.8 out of 100 on a creativity support index and 47.3 out of 100 on task
workload — acknowledging the usefulness of AI in their workflows [2]. At the same time, participants
noted that although AI-generated suggestions were meaningful and contextually relevant, they still
required fine-tuning based on domain-specific knowledge bases [2].
In conclusion, while the potential of AI in design generation is considerable, its implementation in
practice still faces certain limitations and challenges.
Theoretical Background.
Prompt engineering is the science and craft of constructing prompts —
textual inputs given to an AI model — with the aim of eliciting desired outputs. With the emergence of
large language models (LLMs), prompt engineering has gained particular importance, as precise and
well-formulated instructions can significantly influence the model’s responses [3]. In this approach, we
formulate natural language instructions for the model, guiding it to extract and generate internal
knowledge in a targeted direction [3]. Importantly, prompt engineering does not require altering the
model’s parameters or architecture — it leverages the model’s pre-trained knowledge base [3]. For
example, in the study by Gallega and Sumi, specially constructed prompts were used to obtain design
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
335
suggestions from ChatGPT regarding interior materials, resulting in relevant and useful
recommendations [2]. Well-structured prompts unlock the model’s full potential, while vague or
poorly phrased prompts may lead to subpar outcomes.
Zero-shot, one-shot, and few-shot prompting are strategies that differ in the number of examples
provided to the LLM (or other generative models) to guide its output.
Zero-shot prompting provides only a task description, with no example responses. The model
must rely entirely on its pre-trained knowledge to perform the task [3][4]. For instance, asking
ChatGPT “Give me an idea for a bedroom design” constitutes zero-shot prompting — the model
generates an idea based solely on general knowledge acquired during training. While this approach is
simple and widely used, it may lack accuracy since the model must "infer" how to perform the task on
its own [4].
One-shot prompting involves providing one example in addition to the task description [4].
This helps clarify the expected output. For example: “Design a rector’s office like this: [description of
an example]. Now generate a similar design.” This allows the model to consider stylistic cues from the
sample. In the famous GPT-3 experiments by Brown et al., one-shot prompting often yielded
significantly better results than zero-shot, as it offered context and direction — though still limited in
generalization [3].
Few-shot prompting refers to providing two or more examples. This is also known as in-
context learning, where the model is given several demonstrations and asked to generate a new
response in a similar fashion [4]. For instance, one may present several example interior project
descriptions and then ask for a new design that aligns with the patterns in those examples. Few-shot
prompting often results in more accurate and consistent outputs, as the model adapts to the structure of
the given samples [3]. However, this approach requires careful selection and formatting of examples,
as inappropriate samples may introduce unintended bias or reduce performance [3].
Prompt chaining is a technique used to break down a complex task into a sequence of smaller subtasks,
each guided by a separate prompt [5]. Instead of asking the model to complete a compound instruction
all at once — such as “Translate a text from Spanish to English, extract the facts, and translate them
back into Spanish” — the process is broken down: first reading the Spanish text, then translating, then
identifying key points, and so on [5]. This method, popularized in part by IBM researchers, improves
model control and response quality [5]. In prompt chaining, each step’s output becomes the context for
the next prompt, creating a coherent logical flow throughout the process [5].
Prompt chaining has proven particularly useful in multifaceted tasks such as design. For example, a
user may first ask the model to generate a list of design requirements, then create individual solutions
for each requirement, and finally combine them into a cohesive project proposal. This layered
approach enhances the user's control over model outputs and allows for iterative review and refinement
at each stage [5].
Method Definition
Advantage
Limitation
When to Use
Zero-
shot
The AI model is
given a task without
any examples.
Fast,
simple,
good for general
ideas.
Vague,
lacks
control.
For
open-ended
conceptual
generation.
One-shot A task is given with
one example.
More
targeted
output, greater
control.
May
still
lack
sufficient context.
When
generating
ideas in a specific
format.
Few-shot The model is guided
with 2–5 examples.
Context-aware,
more
accurate
results.
Prompts
become
complex.
When copying a
style or structure is
necessary.
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
336
Prompt
chaining
Each
subsequent
prompt builds on the
previous one.
Iterative control,
refined
visual
outputs.
Time-consuming,
requires
experience.
For complex design
tasks and decision-
making processes.
Table 1. Comparative Analysis of Design Generation Methods and Methodologies
Research Methodology.
This experimental study utilized several advanced artificial intelligence (AI)
tools and platforms. For text generation, the ChatGPT model developed by OpenAI was selected. This
model proved effective in constructing prompts, generating design recommendations, and analyzing
intermediate outputs during prompt chaining processes, as well as developing interactive design
scenarios [12], [13].
Image generation was conducted using the Midjourney platform. Midjourney is a cutting-edge
generative AI system designed to produce high-quality and stylistically rich visual compositions based
on user prompts. It is particularly popular in interior, landscape, and conceptual architectural design
projects [14].
Additionally, the experimental design process was supported by practical platforms such as Planner 5D
and ReRoom AI. Planner 5D is an AI-enhanced design environment that enables users to draw room
layouts in 2D and 3D formats, automatically arrange furniture, and explore multiple design alternatives.
Its built-in tools, such as the Design Generator and Smart Wizard, generate design suggestions
automatically based on user-provided parameters [6].
ReRoom AI, on the other hand, allows users to create photorealistic interior renders in over 20
different design styles based on an uploaded photo or 3D model of an existing space. The platform
prompts users to upload an image of their room and select a desired style, after which it produces a
new interior variant within a short time frame [7]. This tool is particularly well-suited for applying
few-shot and prompt chaining techniques in practice, as it enables users to supply an existing example
and generate a new design through AI assistance [15].
Zero-shot Approach (Bedroom Design)
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
337
Figure 1.
Bedroom interior concept generated by the Midjourney model using the
zero-shot approach.
4.1 Zero-shot Approach (Bedroom Design)
In this experiment, no examples were included in the prompt — the model generated a design based
solely
on
the
textual
description
provided
(Figure
1).
To test the zero-shot approach, bedroom design was selected as the first experimental task. Using
ChatGPT, a concise prompt was formulated to define the design requirements, such as: “A 20 sq.m.
bedroom, bright and modern style, with a large window and blue accent details.”
This description, without any visual reference or sample, was then submitted to the Midjourney model.
The resulting images (Figure 1) depict an interior concept generated solely based on the model’s
internal knowledge. That is, Midjourney drew upon thousands of bedroom designs and visual patterns
it had previously encountered during training to synthesize a new and original variation.
The key advantage of this approach lies in its simplicity — minimal input is required from the user: a
clear description is enough. For example, in the study by Gallega and Sumi, designers provided their
requirements to ChatGPT without examples, and the model successfully generated useful textual
recommendations
—
a
classic
case
of
zero-shot
interaction
[2].
In our test, Midjourney’s visual output was of notably high quality. The model independently resolved
aspects such as lighting, furniture layout, and harmony of wall and floor colors. However, the
limitation of this method is that the outcome is less predictable: the user has limited control over the
specific style or focus the model may choose. In fact, zero-shot prompting is often an exploratory and
iterative technique — users may need to generate multiple versions and refine prompts with constraints
in subsequent iterations [4].
In our trial, the first image aligned well with expectations: modern and bright, yet with certain
unexpected decorative interpretations not specified in the prompt — a product of the model’s “creative
inference.”
Overall, the zero-shot approach represents the simplest form of design generation, allowing rapid
exploration of multiple ideas. However, the alignment between output and user intent may be weaker,
due to the absence of guiding examples. As we will see in the next section, including even a single
example (one-shot) improves output precision significantly.
One-shot Approach (Rector’s Office Design)
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
338
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
339
Figure 2.
Office interior generated using the one-shot approach.
In this experiment, a sample layout was provided to the model as a reference for generating a design of
a "rector’s office." As a result, the AI generated a formal and modern office interior aligned with the
sample's style (Figure 2).
As the second test, the one-shot prompting method was applied. The design task involved creating an
office space for a university rector — a setting that demands a specific formal style and high
functionality. To formulate the prompt, ChatGPT was first used to compile a list of design
requirements for a rector’s office (e.g., “Small office size, classic-style furniture, bookshelf units, a
meeting table, and windows allowing good natural lighting”).
According to the one-shot methodology, a single reference example was then added. This included a
description such as: “Example: A classic office with dark wooden furniture and green walls.”
Additionally, Midjourney’s image prompt feature enabled us to upload a small photo of an existing
rector’s office as a visual sample. Once the task description and one visual/textual example were
submitted, the model generated the output shown in Figure 2.
The analysis of this one-shot approach revealed several improvements over zero-shot prompting.
Firstly, the design style was more consistent: classic stylistic features shown in the example — such as
wooden tones and a formal ambiance — were clearly reflected in the generated result. While the zero-
shot approach gave the model full creative freedom, the one-shot prompt guided it to follow the
reference. As highlighted in the literature, providing even a single example can significantly improve
the model’s task understanding [4]. For instance, Brown et al. (2020) demonstrated that GPT-3’s
outputs became substantially more accurate with one-shot prompting [3]. Our visual experiment
confirmed this observation: the office interior generated in one-shot mode better met the functional and
aesthetic expectations.
However, the one-shot method has its limitations — the model adapts strictly to the single given
example. If the reference is of low quality or deviates from the intended direction, the model’s output
may follow that undesired trajectory. In our case, the sample office was well-chosen, but had we
provided a highly futuristic office example, the result could have diverged stylistically. Thus, one-shot
prompting requires careful and skillful example selection by the user.
In summary, one-shot prompting offers the user greater control compared to zero-shot and improves
the alignment between output and intent. This method allows the designer to “teach” the model a
preferred style or reference, shaping the generation process accordingly. However, a single example
may not always suffice — especially when the task is complex or has multiple layered requirements.
Few-shot + Platform Integration (Planner 5D + ReRoom)
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
340
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
341
Figure 3.
3D visualizations created in Planner 5D based on the layout.
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
342
Figure 4. ReRoom AI interface.
Figure 5.
Interior design stages generated using ReRoom AI.
In the few-shot approach, the model is guided through several examples before generating the final
design output via an AI-powered platform.
In the third experiment, the few-shot method was integrated with specialized design platforms. The
task remained the same as before: to generate an interior design for a university rector’s office. First,
the office floor plan, originally drafted in AutoCAD, was uploaded into
Planner 5D
(Figure 3). The
software then produced a 3D visualization of an empty room layout (Figure 4). These 3D
visualizations were subsequently uploaded into
ReRoom AI
for style transformation.
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
343
In ReRoom AI, the few-shot process unfolded as follows: a 3D model of the interior was provided as
input, and then the user selected a new target style from the platform's interface [8]. For instance, we
asked the system to transform a “loft-style” interior into a “Scandinavian” style. This was achieved by
selecting the corresponding style option from ReRoom’s menu (Figure 4). Internally, ReRoom likely
uses Generative Adversarial Networks (GANs) or Diffusion Models to analyze the original image and
generate a stylistically transformed render within seconds. The outcome of this transformation is
shown in Figure 5.
So where is the few-shot element here?
First, multiple design variants produced in Planner 5D can serve as examples for ReRoom AI — in our
specific test, a single 3D layout was selected, but the platform had previously been tested with other
styles as well.
Second, ReRoom AI is pre-trained on more than 20 interior design styles, meaning that the model has
already learned from multiple examples per style [8]. When a user selects “Scandinavian,” for example,
the platform leverages numerous embedded references to generate the new result. Although the
process appears seamless to the user, internally it draws upon a rich set of style-specific data — a
modeled form of few-shot learning.
The results of this experiment demonstrated that the few-shot approach produced the most accurate
and user-aligned outcomes. For instance, when the room was transformed into a Scandinavian style,
the model replaced the exposed brick wall texture with white paint and wooden panels, and substituted
furniture with simple, light-colored alternatives — all typical of Scandinavian interiors. Achieving
such specific stylistic shifts through zero-shot or one-shot methods would have been significantly more
difficult, as these would require detailed textual explanations that the model may or may not interpret
accurately. In contrast, few-shot prompting allows the model to learn from context, resulting in more
reliable outputs [3].
Scientific literature also supports this: LLMs trained with multiple examples tend to commit fewer
errors and adapt more precisely to new instructions [3]. Our visual experiment confirmed this claim.
Of course, the few-shot + platform method is not without its challenges. This approach requires the
most user involvement: the user must first create or select several variants, upload them to the platform,
and then configure the necessary style parameters. In short, the process is more time- and attention-
intensive.
However, in terms of quality, this method produced the highest-rated outputs — the generated interiors
were stylistically coherent, rich in detail, and well-aligned with functional requirements. The degree of
user control was also significantly higher, as designers could influence the process at multiple stages
(creating variants, selecting the best option, and defining the final style).
In conclusion, the few-shot approach — especially when integrated with specialized platforms —
delivers highly effective results in design generation. It mirrors real-world design workflows, where
designers often review several options before selecting a final solution. AI-enhanced few-shot
techniques are well-suited to accelerating this process in professional design contexts.
Prompt chaining
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
344
In this example, the task of designing a recreational landscape for the surroundings of the Sho‘rko‘l
Reservoir, located in Bukhara, was carried out step-by-step with the support of artificial intelligence.
Unlike previous zero-shot experiments, this scenario involved describing a complex design problem
through a sequence of prompts — a methodology known as
prompt chaining
(Figure 6).
Step 1: Defining Requirements and Establishing Context
The first prompt given to ChatGPT was:
“List the key geographic and social factors to consider when designing the landscape around the
Sho‘rko‘l Reservoir.”
The model recommended the following considerations:
Ecological and topographic features near the water (e.g., sandy shores, lowland canal networks);
Figure 6.
Landscape design concept generated using the prompt chaining approach.
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
345
Potential for recreation, sports, and eco-tourism (e.g., beach zones, walking and running paths,
playgrounds, sports areas);
Integration of traditional and modern architecture (e.g., towers, mausoleum-style buildings,
open-air stage);
Visual harmony (e.g., circular and biomorphic forms, colorful walkways).
These recommendations were reviewed, and the most relevant ones were retained to match the site's
location and needs.
Step 2: Developing Compositional Concepts
Next, we asked the model:
“Based on the criteria above, generate a conceptual landscape design.”
The AI proposed the following compositional elements:
A traditional domed building (e.g., historical madrasa or cultural center) as the central anchor;
Symmetrically arranged walkways and oriental-style gardens surrounding it;
Elongated recreational zones along the beach: shaded areas, picnic spaces, and an open-air
amphitheater;
A large water feature (fountain or canal) with surrounding circular colored pathways.
Step 3: Expanding the Concept with Functional Additions
The subsequent prompt requested:
“Add features like water slides, sports areas, and residential cottages to the above design.”
ChatGPT responded with:
Various types of aquapark slides;
Flat-roof or brick-roof vacation cottages;
Aesthetically integrated football, basketball, and tennis courts;
Decorative circular landscaping elements.
Step 4: Refining Details, Textures, and Lighting
To further enhance the visualization, a final prompt was submitted:
“Mark the pathways with colorful pavement, add more palm trees, include sunset lighting effects, and
render the view from a drone-like aerial perspective.”
This final step produced a highly realistic composition, capturing natural lighting, shading, and overall
visual balance.
This experiment demonstrated that working in sequential prompt chaining mode — rather than giving
a single general prompt — significantly improved the quality of visual results. Each new prompt
served as a logical continuation of the previous one, enabling conceptual evolution and refinement.
The user's inputs were gradually refined based on social, cultural, and technical criteria, resulting in a
coherent, adaptive, and visually compelling landscape design.
Prompt chaining not only enhances visual output quality but also supports the development of design
thinking, models decision-making sequences, and activates the creative potential of artificial
intelligence.
Summary of Findings
. The four experiments conducted demonstrate that each AI-based design
generation approach has its unique strengths and limitations. They can be comparatively evaluated
based on the following key criteria:
Visual quality (accuracy and relevance),
Flexibility (ability to vary in style and concept),
Control level (extent of user influence over the process),
Prompt sensitivity (output responsiveness to input),
User involvement (level of creative and functional input required).
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
346
Eksperim
ent usuli
Vizual sifat
(aniqlik,
moslik)
Moslashuvch
anlik
Nazorat
imkoniyati
Promptga
sezgirlik
Foydalanuv
chi ishtiroki
Zero-shot
Average: low
precision in
overall
compositions
Low:
ideas
remain vague
Very low: user
cannot predict
results
Moderate: output
is broadly related
to the prompt
Minimal:
only initial
prompt
input
One-shot
Good: clear
form
and
concept
Moderate:
limited to one
example
Moderate:
some control
within narrow
scope
High:
accurate
response
to
specific
prompt
details
Medium:
based on a
single input
Few-shot
High: strong
coherence in
style,
tone,
and form
Good: guided
by
multiple
examples
Moderate:
adaptable but
pre-defined
High:
deep
alignment
with
prompt
and
context
Active: user
provides
structured
context
Prompt
chaining
Very
high:
increasing
accuracy at
each step
Very
high:
editable at any
stage
High:
user
influences
each step
Very high: highly
responsive
to
each prompt
Maximum:
ongoing
interaction,
refinement,
and analysis
Table 2. Analysis of Experimental Approaches
Conclusion.
This analysis demonstrates that the prompt chaining method provides the highest level of
user engagement and control. As a result, it serves as the most suitable approach for developing
complex and conceptually rich design projects. In contrast, the zero-shot approach remains the least
controllable but is highly efficient for rapid prototyping. The one-shot and few-shot methods present
more balanced alternatives, particularly effective for generating stylistically coherent design solutions.
In this study, four methodological approaches to AI-driven design generation—zero-shot, one-shot,
few-shot, and prompt chaining—were explored through experimental trials. Each method was
evaluated individually, and the results were compared based on criteria such as visual quality, user
control, creative supervision, complexity, and interactivity.
The zero-shot approach, while offering quick conceptual outputs, proved limited in accuracy and user
control. The one-shot and few-shot methods allowed for greater contextual and stylistic precision by
incorporating examples, but they exhibited constraints in prompt sensitivity. Among all, prompt
chaining emerged as the most effective method. It enabled users to participate in each stage of the
process, allowing for iterative refinement and structured decision-making in complex design
development.
The landscape design case study at the Sho‘rko‘l reservoir demonstrated that advanced tools such as
ChatGPT, Midjourney, Planner 5D, and ReRoom AI can be successfully integrated through an
iterative prompt chaining approach to yield high-quality conceptual and visual design results. This
approach has proven effective in maintaining a balance between AI automation and user-driven
creativity in the design workflow.
INTERNATIONAL CONFERENCE
Dehli, India, 2025
https://eijmr.org/conferences/
INNOVATIONS IN SCIENCE AND EDUCATION SYSTEM
347
Based on these findings, the prompt chaining methodology is proposed as one of the most promising
approaches for establishing a collaborative, flexible, and creative interaction between designers and
artificial intelligence systems. It holds particular potential for addressing complex, multi-stage
challenges in contemporary architectural and design practices.
References
1. Huang, J., Bibri, S.E., Keel, P. (2025). Generative Spatial Artificial Intelligence for Sustainable
Smart Cities: A Pioneering Large Flow Model for Urban Digital Twin. (Manba: ResearchGate)
2. Gallega, R.W. & Sumi, Y. (2024). Exploring the use of generative AI for material texturing in 3D
interior design spaces. Frontiers in Computer Science, 6:1493937
3. Vatsal, S. & Dubey, H. (2024). A Survey of Prompt Engineering Methods in Large Language
Models for NLP Tasks. arXiv preprint arXiv:2407.12994
4. Prompting Guide (2025). Shot-Based Prompting: Zero-Shot, One-Shot, and Few-Shot Prompting.
(online qo‘llanma)
5. IBM (2024). What is prompt chaining? IBM Technology Blog, 23-aprel 2024
6. Planner5D. Artificial Intelligence in Interior Design. URL:
https://planner5d.com/use/ai-interior-
7.
Deepgram
. ReRoom: AI-Powered Interior Design Visualizer. URL:
8.
ReRoom AI
. Interyer dizayni uchun sun'iy intellekt vositasi. Elektron resurs. URL:
. Murojaat qilingan sana: 28.06.2025.
9. Cai, A. va boshq. (2023). DesignAID: Using Generative AI and Semantic Diversity for Design
Inspiration. UIST 2023. (DesignAID tizimida prompt engineering va few-shot usullarining
qo‘llanilishi bayon etilgan)
10.
Anthropic
.
Chain
complex
prompts
for
stronger
performance.
URL:
https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/chain-prompts
11.
Botpress
. What is AI Prompt Chaining? URL:
https://botpress.com/blog/what-is-ai-prompt-
chaining#:~:text=What%20is%20AI%20Prompt%20Chaining%3F,step%20task
12. Brown, T. et al. (2020). Language models are few-shot learners. NeurIPS.
13. Zhao, X. et al. (2023). Prompt Chaining in Generative Design Workflows. CHI Conference.
14. Burry, M. (2021). Designing with AI. Automation in Construction.
15. Chen, J. et al. (2023). Zero-shot generative interior design. Journal of Computational Design.
16. Rakhmatillaeva Z. Z., Vetlugina A.V., (2023).
Patterns and principles of constructing a
landscape composition
. Journal of Modern Educational Achievements, 5(5), 216–224.
17. Rakhmatillaeva, Z. Z. (2020). The use of natural monuments of Surkhandarya region for
ecological
tourism.
file].
Retrieved
from
idpublications.org
website:
https://www.idpublications.org/wp-content/uploads/2020/12/Full-Paper-THE-USE-OF-
