Development and applications of Generative AI in architectural design studios

Synthesis: Examines the integration of deep generative models into architectural design education. The findings, based on students' views and observations in design studios, suggest that GenAI supports the exploration of creative ideas — serving as visual stimuli and inspirational resources in early design stages — while also highlighting the competencies students need to differentiate between GenAI models and use them effectively.

Key Findings

• A longitudinal study in architectural design studios led to the development of the GAI-A platform, a user-friendly GenAI interface refined through a cyclical process of experimentation, feedback collection, and evaluation.
• Students used GenAI as visual stimuli and inspirational resources, supporting exploration of the unknown in the early stages of the design process.
• A second, combinatorial pattern of use (analogous to Boden's 'combinatorial' method) was effective as solution space enhancement: generating alternatives that expand the solution space during the design and development phase, with students using GenAI to advance their preliminary solution ideas.
• The authors stress that avoiding design fixation requires competencies and discernment to differentiate between GenAI models and use them effectively.
• The study also cites evidence (Wadinambiarachchi et al., 2024) that exposure to GenAI images can narrow designers' focus by tying them to specific aesthetics, and notes that several of its own findings require further research.

Study Design & Method

The research combined the development of a GenAI interface with its implementation in design education: GenAI models were used in design studios, feedback was collected, the GAI-A platform was developed and evaluated, and it was subsequently implemented in design education. Findings draw on students' views and observations, including semi-structured interviews with a focus group conducted in week 6 (W6) and week 13 (W13) of studio work, with coded design processes illustrating how GenAI entered the design workflow.

What this means for practice

• Instructors. Teach model selection and discernment alongside generation. The study's central claim is that avoiding design fixation depends on competencies to differentiate between GenAI models and use them effectively, so many short comparison exercises with different models serve students better than proficiency with one.
• Instructors. Sequence GenAI to the design stage rather than treating it as one uniform tool: in these studios students used it as visual stimuli and inspiration while exploring the unknown early on, as solution-space enhancement when developing preliminary ideas, and as externalization support during documentation.
• Learners. Guard against aesthetic lock-in. Exposure to GenAI images can narrow focus by tying designers to specific aesthetics (Wadinambiarachchi et al., 2024), and the models in this studio showed a tendency to generate images of comparable style.
• Learners. Build a personal style repository and prompt with it. The authors propose collaborating with GenAI using the designer's own collection of styles as the more promising route to an original architectural language.
• Instructors. Make delegation an explicit object of studio critique, since students in this study valued the time savings while remaining hesitant to relinquish control of design decisions.

Limitations

• The qualitative evidence rests on focus-group interviews with six participants (N = 6), each roughly 30 minutes, conducted at week 6 and week 13 of one part of a multi-semester study; the authors state that the small sample and the participants' limited experience preclude broad generalization.
• Survey data came from a single cohort in the third semester (Case-III: 24 students enrolled, N = 18 at week 6 and N = 24 at week 13), and the study reports students' views and studio observations rather than a controlled comparison, so the documented patterns cannot establish that GenAI caused the reported creative gains.
• Creativity itself is inherently subjective and resists objective evaluation, which the authors name as a central limitation: what one observer judges creative another may find banal, and the coded design processes map reported steps rather than measured originality.
• Several findings are flagged by the authors as needing further research, including the claim that relinquishing control to GenAI establishes a new focal point for exploring creativity.

Citation

Leman Figen Gül, Burak Delikanlı, Oğulcan Üneşi, Ertuğrul Ömer Gül (2026). Development and applications of Generative AI in architectural design studios.