ITU GAI-A

GAI-A : Generative Artificial Intelligence for Architecture

An AI-assisted co-design interface for the early stages of architectural design

Input: a student's collage
Output: generated in GAI-A
project
Enhancing creativity through AI collaboration: What is the role of Generative AI in architectural design education?
institution
Istanbul Technical University, Faculty of Architecture
funding
ITU BAP, project MGA-2023-44787
tested in
Third-year architectural and interior design studios
duration
2022–2025, completed

Project

ITU BAP project MGA-2023-44787

Enhancing creativity through AI collaboration: What is the role of Generative AI in architectural design education?

Image-generation models reached architecture students in 2022, long before studio teaching had a way to work with them. This ITU BAP project set out to find where these models help in the design process. Along the way it produced GAI-A, a tool that lets students use them without programming.

The project followed students across several semesters of design studios at the ITU Faculty of Architecture. Each semester built on what the previous one taught us: we observed how students used generative models, asked what they needed, and turned those needs into the GAI-A interface. The interface was then tested in the studio and evaluated with surveys, semi-structured interviews and the Creativity Support Index.

Research questions

  • How do design students bring Generative AI models into their design process, and at which stages?
  • How can these models support the different stages of design, and what should a user-friendly interface for them look like?
  • What benefits do students see in Generative AI, and how far do they believe it supports their creativity?

The studios and studies behind these questions are listed under Studio series.

Findings

ResultWhat was measuredSource
95%of students said the Generative AI models they used had a positive effect on their creativity, in week 6 of the semester (N=18)A|Z 2025
96%said the same at the end of the semester, in week 13 (N=24)A|Z 2025
87%of students saw GAI-A's influence on their creativity as positive (N=18)CDVE 2024
89%rated GAI-A as a helpful partner for producing a variety of design ideas and alternativesCDVE 2024
84%said the generated images helped them conceptualise design ideas and widened their creative horizonsCDVE 2024
75.75Creativity Support Index score for GAI-A, out of 100, in a workshop with third-year students (N=18). Enjoyment and results worth effort scored highest; immersion scored lowest.CDVE 2024
eCAADe 2025
95.3Creativity Support Index score in an expert study with a doctoral researcher in architectural design computingeCAADe 2025

Students found Generative AI models most useful in the uncertain, early phases of design, for gathering inspiration and making abstract ideas visible. Their use dropped off as projects moved towards a defined solution, and many students preferred to rework generated images to keep authorship of their designs. A|Z 2025

from the studios, 2022–2024Four ways students worked with Generative AI

Observations across the studios showed four approaches, described in MSTAS 2024. Across all semesters, students' command of the tools was the clearest predictor of how far they could take their design process.

Exploring the unknown
Using generated images for inspiration and concept development in the earliest phase, especially when the brief or program was still undefined.
Expanding the options
Testing many alternatives during design development. Students who moved fluidly between tools and fed results back into their process worked most productively.
Representing the known
Visualising decisions already made. This saved time, but the models tend to repeat popular images, which limits originality.
Developing a method
Putting language models and prompting at the centre of the process, working out ideas in words before turning them into images.

Interface

GAI-A puts Stable Diffusion workflows behind one simple screen. Students upload and arrange images, write a prompt and generate, without installing software or building node graphs.

Basic fields are shown by default. An advanced mode in each workflow opens nearly all of its internal parameters for experienced users.

Workflows

Blend img2img · denoise
Blends a collage of input images into a single image. The Denoise setting controls how closely the result follows the collage, from literal to loosely abstract.
Canny controlnet · canny
Traces the edges of an input image and keeps those contours while generating new materials, textures and detail. Suited to sketches and line drawings.
Depth controlnet · depth
Estimates the depth of an input image so the result keeps its spatial relationships and perspective. Suited to model views and existing renders.
Text txt2img
Generates an image purely from text prompts.

How it works

01Choose a workflow
Switch between Blend, Canny, Depth and Text with the toggles on the canvas.
02Build the input
Drop images into the workspace frame, then resize, rotate and reorder them.
03Write prompts
Describe what you want, and in the negative prompt what to avoid.
04Generate
The workflow turns noise into an image. Tick Iterate to keep the seed and make small changes to a result.
05Review outputs
The Outputs tab keeps your generation history, so you can compare and save results.

Read the user manual for each workflow and setting in detail.

How it is built

  • Web applicationA Next.js interface hosted on Vercel: the canvas, prompts, workflow switch and outputs.
  • DeploymentComfy Deploy turns each workflow into an online API and runs the models, so nothing runs on the student's computer.
  • GenerationComfyUI workflows built on Stable Diffusion, with edge and depth conditioning.

Three prototypes

  1. Collage toolTurned a collage into a single render: Midjourney described the collage, ChatGPT refined the prompt, Midjourney generated the image.
  2. Stable DiffusionMoved to open models for more control. Added edge and depth tools, an abstraction setting, advanced options, layer ordering and same-seed generation.
  3. One interfaceBrought all tools onto one canvas with workflow toggles, and added a text-only workflow.

Studio series

GAI-A grew out of a research-driven studio series. Since 2022, successive third-year architectural design studios at ITU have asked students to work with Generative AI, each building on what the previous one taught us.

Following successive cohorts of architecture students lets us see how their use of these tools changes as the models, the interface and the teaching change around them.

  1. I – Fall 2022

    Designing the Black Box: Creating the Intimate Space of Arts

    Cibali, Istanbul22 students

    Studio team: Prof. Dr. Leman Figen Gül, Res. Asst. Burak Delikanlı

    ToolsChatGPTDALL-E 2Midjourney

    Spaces for art events, balancing public interface, assembly and performance. The studio opened with an introductory workshop on ChatGPT, DALL-E 2 and Midjourney, before any of these tools were common in teaching.

  2. II – Spring 2023

    Neo-Education: Designing the Future Education Spaces

    Germencik, Aydın18 students

    Studio team: Prof. Dr. Leman Figen Gül, Res. Asst. Burak Delikanlı

    Guest Lec. Begüm Moralıoğlu

    ToolsChatGPTDALL-E 2MidjourneyMicrosoft HoloLensMeta Quest 3

    Future learning environments in an educational complex. A five-session workshop combined spatial computing (AR and VR) with Generative AI models to imagine spaces that did not yet exist.

  3. III – Fall 2023

    Envisioning Future Urban Life: Infill Design in a Historical Urban Setting

    Kemeraltı, İzmir26 students

    Studio team: Prof. Dr. Leman Figen Gül, Res. Asst. Burak Delikanlı

    Guest Lecs. Salih Özdemir, Selen Çiçek, Merve Akdoğan, Oğulcan Üneşi

    ToolsChatGPTDALL-E 3MidjourneyStable DiffusionComfyUIControlNetGAI-A

    Community hotspots in an urban heritage setting, approached through design fiction. Seminars and six workshops covered Stable Diffusion, ControlNets and ComfyUI, and the GAI-A Blend tool was tested in the studio for the first time. Surveys and interviews from this semester form the core of the published findings.

    Output: CDVE 2024, eCAADe 2025, A|Z 2025

  4. IV – Spring 2024

    Theatrum Deus Ex: Orchestrating the Symphony of City Life

    Bağdat Caddesi, Istanbul21 students

    Studio team: Prof. Dr. Leman Figen Gül, Res. Asst. Burak Delikanlı

    Guest Lecs. Salih Özdemir, Oğulcan Üneşi

    ToolsChatGPTDALL-E 3MidjourneyStable DiffusionComfyUIControlNetGAI-A

    Art and culture spaces that shape the daily life of the city. In the first four weeks students worked with language and diffusion models, including the GAI-A interface, through a seminar and three workshops, then combined their outputs into a narrative flow diagram for their projects.

    Output: MSTAS 2024

  5. Study – Summer 2024

    Protocol study with GAI-A

    ITU, with 1 expert designer and 30 architecture students

    ToolsGAI-A

    Recorded design sessions in which participants worked with GAI-A, coded and analysed to understand how designers use a Generative AI companion and where it supports their creativity.

    Output: Journal article (in preparation)

  6. V – Fall 2024

    Center for Research & Education for Ecological Production of Food

    Eğirdir, Isparta33 students

    Studio team: Prof. Dr. Leman Figen Gül, Res. Asst. Burak Delikanlı

    Guest Lec. Senem Özden Gerçeker

    ToolsChatGPTStable DiffusionAutodesk FormaVeras

    Research and education spaces for ecological farming and food production in a rural setting, guided by green building and circular design principles. This semester the AI focus moved to AEC tools: students used Autodesk Forma for site and environmental analysis and Veras for AI rendering in their presentations.

  7. VI – Spring 2025

    Nourish & Flourish: Fener-Balat Culinary and Community Engagement

    Fener-Balat, Istanbul25 students

    Studio team: Prof. Dr. Leman Figen Gül, Res. Asst. Burak Delikanlı

    ToolsChatGPTStable DiffusionAutodesk FormaVeras

    Gastronomy, short-term vocational training for women and youth, and accommodation for trainees, designed to revitalise a historic neighbourhood. Students used ChatGPT to speculate about activities and spaces, and Autodesk Forma and Veras for analysis and AI rendering.

  8. VII – Fall 2025

    ReCoding the City: Regeneration Through Connection

    Üsküdar, Kadıköy or Şişli, Istanbul23 students

    Studio team: Prof. Dr. Leman Figen Gül, Lec. Sinem Kırkan

    ToolsChatGPTClaudeStable Diffusion

    Housing and urban regeneration in Istanbul, working through real cases of urban transformation and their social, legal and economic sides. An LLM and Generative AI seminar on text-to-text, text-to-image and image-to-image models and prompting fed the concept phase, alongside ChatGPT, Claude and Stable Diffusion.

    Output: ICNTAD 2026, Book chapter (forthcoming)

  9. VIII – Spring 2026

    Urban Infill for Cultural Encounter

    Karaköy, Cihangir, Şişli or Beyoğlu, Istanbul13 students

    Studio team: Prof. Dr. Leman Figen Gül, Lec. Sinem Kırkan

    Guest Lec. Burak Delikanlı

    ToolsChatGPTClaudeStable DiffusionComfyUIControlNet

    A boutique hotel that bridges Istanbul’s historical identity and contemporary hospitality. Alongside ChatGPT and Claude, students worked with node-based multimodal Generative AI workflows in both problem-search and solution-space tasks.

    Output: Journal article (in preparation)

  10. IX – Fall 2026

    Hybrid Work-Live Interfaces along the Büyükdere Axis

    Levent, Istanbul20 students

    Studio team: Prof. Dr. Leman Figen Gül, Res. Asst. Burak Delikanlı

    Guest Lec. Murat Gül

    ToolsChatGPTClaudeAutodesk FormaVeras

    Permeable, multi-layered interfaces for living and working in the high-rise corridor of Istanbul’s financial district, set against its monolithic, privatised towers. Students work with language and image models, approaching the project as non-linear human–AI co-creation. In progress.

From the studios

A selection of the inputs students brought and the images they worked with.

Fall 2023Sude Gurur YenicanEnvisioning Future Urban Life, KemeraltıBlend workflow: an early-stage collage of a massing model, references, people and trees turned into one design image.
Spring 2024Atacan KüçüksarıTheatrum Deus Ex, Bağdat CaddesiDepth workflow: a hand sketch over a city backdrop used to develop the facade and massing in the early design phase.
Summer 2024Emir ÇiçekProtocol study with GAI-AA sketch uploaded to GAI-A to explore materials and facade: the hand-drawn massing returns as a concrete building.

Publications

  1. Article in preparation, expected 2027

    Node-based multimodal Generative AI in the architectural design studio: An exploratory study of student perceptions across problem-search and solution-space tasks

    B. Delikanlı, S. Kırkan, L. F. Gül.

    A semester-long mixed-methods study of architecture students using a transparent, node-based Generative AI environment (ComfyUI). Students’ attitudes recalibrated rather than swinging to enthusiasm or rejection: task-level usefulness held steady, expectations that AI could solve complex problems cooled, and the felt need to verify outputs stayed high.

  2. Article in preparation, expected 2027

    A protocol study of revealing creative potentials of Generative AI companion in architectural design

    L. F. Gül, B. Delikanlı, S. Kırkan.

    A protocol study with one expert designer and 30 architecture students working with GAI-A. It examines how designers engage with a Generative AI companion while they design, and where that collaboration opens up creative potential.

  3. Book chapter, forthcoming

    A rhizomatic framework for Generative AI in architectural design studios

    L. F. Gül, D. Arslan Dinçay, S. Kırkan, B. Delikanlı, Y. Çoraklı. In O. Z. Güzelci and İ. As (Eds.), The Routledge Companion to Artificial Intelligence in Architecture, 2nd ed. Routledge.

    Describes a studio that combines rhizomatic learning with Generative AI through four modules: Data Ignition, Branching, Collective Tectonic Resolution and Rhizomatic Mapping. Peer learning, ethical awareness and atmospheric exploration flourished, but students’ design judgement remained susceptible to the authority of AI-generated images, which calls for structured pedagogical scaffolding.

  4. Conference paper, 2026

    A cohort-sequential study on student perceptions of designing with Generative AI in design studios

    D. Arslan Dinçay, L. F. Gül, S. Kırkan, Y. Çoraklı, B. Delikanlı. 12th International Conference on New Trends in Architecture and Interior Design (ICNTAD’26), Barcelona.

    Compares how successive cohorts of design students perceive designing with Generative AI in the studio.

  5. Journal article, 2025

    Development and applications of Generative AI in architectural design studios

    L. F. Gül, B. Delikanlı, O. Üneşi, E. Ö. Gül. A|Z ITU Journal of the Faculty of Architecture, 22(3), 751–774.

    A longitudinal study across successive design studios: how students bring Generative AI into their design process, how the GAI-A interface grew out of their needs, and how far students believe these tools support their creativity.

  6. Conference paper, 2025

    User-centered Generative AI framework for visual stimuli in architectural design: Development and evaluation of the GAI-A interface

    L. F. Gül, O. Üneşi, B. Delikanlı, E. Ö. Gül. Confluence: Proceedings of the 43rd eCAADe Conference, Vol. 1, 461–470.

    Describes the three layers of the GAI-A interface (Stable Diffusion workflows in ComfyUI, deployment with Comfy Deploy, and the web application) and evaluates it with the Creativity Support Index in an expert study and a student workshop.

  7. Conference paper, 2024

    Exploring co-design with an AI partner: The GAI-A interface in architectural education

    L. F. Gül, B. Delikanlı, O. Üneşi, E. Ö. Gül. In Y. Luo (Ed.), Cooperative Design, Visualization, and Engineering (CDVE 2024), LNCS 15158, 1–12. Springer.

    Introduces GAI-A as a co-design partner for the early design phase and reports its evaluation in the architectural design studio through the Creativity Support Index, surveys and semi-structured interviews.

  8. Conference paper, 2024 (in Turkish)

    Employing Generative AI models in architectural design education: Our insights from the studio experience

    L. F. Gül, B. Delikanlı, O. Üneşi, E. Ö. Gül. XVIII. National Symposium on Digital Design in Architecture (MSTAS 2024), Balıkesir University.

    Reflects on four semesters of studio work with Generative AI and identifies four ways students used it: exploring the unknown, expanding the options, representing the known and developing a method.

Team

  • Leman Figen Gül

    Prof. Dr. Leman Figen Gül

    Principal Investigator

    Istanbul Technical University, Department of Architecture

    Dr. Leman Figen Gül is a Professor of Architecture at Istanbul Technical University (ITU), where she coordinates the Architectural Design Computing Graduate Program and chairs the Human Research Ethics Board in Social and Human Sciences. She previously headed the Game and Interaction Technologies Program (2020–2025) and directed the ITU Women's Studies Center in Science, Engineering, and Technology (2018–2021). She earned her PhD in Architecture from the University of Sydney’s Key Centre of Design Computing and Cognition (2007) and has held academic and research positions at the University of Newcastle, International University of Sarajevo, and TOBB ETU. Dr. Gül has led numerous national and international research projects funded by TÜBİTAK, Erasmus+, and various research councils. Her interdisciplinary research focuses on design computing, Generative AI in design pedagogy, immersive technologies, spatial cognition, and gamified cultural heritage interpretation.

  • Demet Arslan Dinçay

    Assoc. Prof. Dr. Demet Arslan Dinçay

    Associate Professor

    Istanbul Technical University, Department of Interior Architecture

    Dr. Demet Arslan Dinçay is an Associate Professor of Interior Architecture at Istanbul Technical University (ITU) and the coordinator of the International Master of Interior Architectural Design (IMIAD) Program. She received her Ph.D. in Architectural History from Istanbul Technical University and has been a visiting professor at Auburn University, USA. Her research focuses on interior design and interdisciplinary approaches to the field, with particular interests in Generative AI in interior design education, virtual and cinematic spaces, and emerging technologies in design practice and pedagogy. She coordinated an interdisciplinary EU-funded project on the use of serious games in interior design education and has participated in several research projects supported by national research institutions. She has also edited books focusing on extended realities, virtual environments, interactive experiences, game-based learning, and design education.

  • Burak Delikanlı

    Burak Delikanlı

    Research Assistant

    Istanbul Technical University, Department of Architecture

    Burak Delikanlı is a Research Assistant in the Department of Architecture at Istanbul Technical University (ITU), where he is currently a PhD Candidate in the Architectural Design Computing Graduate Program. His research interests include Generative AI in design, design creativity, computational design, BIM, and digital fabrication, with a particular focus on Human-AI collaboration in design processes. He has taken part as a researcher in national and international research projects funded by TÜBİTAK and Erasmus+. His work combines research, teaching, and hands-on prototyping, and he is interested in translating emerging computational methods into usable tools, workflows, and educational settings.

  • Oğulcan Üneşi

    Oğulcan Üneşi

    PhD Candidate

    Istanbul Technical University, Architectural Design Computing Graduate Program

    Oğulcan Üneşi is a PhD candidate in the Architectural Design Computing Graduate Program at Istanbul Technical University. He focuses on computational design, artificial intelligence, programming and software development for architecture. He has tutored in design geometry, computational design and basic design courses, and has undertaken research and programming roles in scientific research projects. Alongside his academic work, he supports research institutions and architectural offices in computational design and software development. He received his Bachelor's degree in Architecture from Mimar Sinan Fine Arts University in 2017, a Master's degree in Advanced Architecture from the Institute for Advanced Architecture of Catalonia in 2018, and a second Master's degree in History, Theory, and Criticism of Architecture from Istanbul Bilgi University in 2022.

  • Ertuğrul Ömer Gül

    Ertuğrul Ömer Gül

    Computer Engineer

    University of New South Wales, School of Computer Science and Engineering

    Ertuğrul Ömer Gül is a software engineer who has held a range of roles in the technology and education sectors. At the University of New South Wales (UNSW) he works as a Casual Academic Tutor and Lab Assistant, guiding students in object-oriented design and programming. Working remotely with Istanbul Technical University, he plays a key role in the full-stack development of an application that brings AI into architectural design and education. He has also worked for Kantar Australia in marketing analytics support, and as a software intern at Ulukom/WiseTech Global Group on security-focused projects. He is studying for a Master's degree in Complex Systems at the University of Sydney, and graduated from UNSW with a double major in Computer Science and Physics.

  • Sinem Kırkan

    Sinem Kırkan

    Lecturer

    Istanbul Technical University, TRNC Education and Research Campus (Famagusta)

    Sinem Kırkan is a Lecturer at the ITU-TRNC Education and Research Campuses. She holds a bachelor’s degree in Architecture and a master’s degree in Architectural Design Computing, both from Istanbul Technical University, where she is currently a PhD Candidate in the Architectural Design Computing Graduate Program. Her academic interests include human–space interaction, spatial cognition, design computing, and the computational modeling of spatial perception. Her research focuses on understanding how individuals perceive, experience, and learn spatial environments, with particular emphasis on the integration of computational methods, virtual environments, and data-driven approaches in the study of human–environment relationships.

  • Yasemin Çoraklı

    Yasemin Çoraklı

    Research Assistant

    Istanbul Technical University, Department of Interior Architecture

    Yasemin Çoraklı is a Research Assistant in the Department of Interior Architecture at Istanbul Technical University. She holds a bachelor’s degree in Architecture from Mimar Sinan Fine Arts University and a master’s degree from the International Master of Interior Architectural Design (IMIAD) program at Istanbul Technical University. She is currently pursuing her PhD in Interior Architecture at Mimar Sinan Fine Arts University. Her academic interests include interior architecture and design education, design studio pedagogy, material–design relationships, and digital approaches to design. Her research and teaching activities focus on design processes, innovative learning approaches, and the integration of contemporary design tools into interior architecture education.

Research Interns

  • Ali Kemal Akbulut

    Ali Kemal Akbulut

    Research Intern (Protocol Study)

    Istanbul Technical University, Department of Architecture

  • Zeynep Ceren Akkaya

    Zeynep Ceren Akkaya

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Ecem Alkan

    Ecem Alkan

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Ahmet Behlül Argat

    Ahmet Behlül Argat

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Elena Asllanaj

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Ege Ayaksız

    Ege Ayaksız

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Ayberk Aydın

    Ayberk Aydın

    Research Intern (Protocol Analysis)

    Istanbul Technical University, Department of Architecture

  • Arman Bayboğan

    Arman Bayboğan

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Emir Çiçek

    Emir Çiçek

    Research Intern (Protocol Study)

    Istanbul Technical University, Department of Architecture

  • Sevban Dağlıoğlu

    Research Intern (Protocol Analysis)

    Istanbul Technical University, Department of Architecture

  • Merve Erdoğan

    Research Intern (Protocol Study)

    Istanbul Technical University, Department of Architecture

  • Selimcan Gök

    Selimcan Gök

    Research Intern (Protocol Study)

    Istanbul Technical University, Department of Architecture

  • Mervenur Gül

    Mervenur Gül

    Research Intern (Protocol Study)

    Istanbul Technical University, Department of Architecture

  • Valeria Halili

    Valeria Halili

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Emin İşçi

    Emin İşçi

    Research Intern (Website Development)

    Istanbul Technical University, Department of Architecture

  • Bediha Kellecioğlu

    Bediha Kellecioğlu

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Atacan Küçüksarı

    Atacan Küçüksarı

    Research Intern (Website Development)

    Istanbul Technical University, Department of Architecture

  • Mahmut Talha Reşitoğlu

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Sinem Subaşı

    Sinem Subaşı

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Faruk Tuğra Şahin

    Research Intern (Protocol Study)

    Istanbul Technical University, Department of Architecture

  • Pelin Tezcan

    Pelin Tezcan

    Research Intern (Protocol Study)

    Istanbul Technical University, Department of Architecture

  • Şeyma Topal

    Şeyma Topal

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Zeynep Öykü Yahşi

    Zeynep Öykü Yahşi

    Research Intern (Literature Review)

    Istanbul Technical University, Department of Architecture

  • Zeynep Berra Zontur

    Zeynep Berra Zontur

    Research Intern (Protocol Analysis)

    Istanbul Technical University, Department of Architecture