AI-Enhanced Visualization in Architectural Education

Why: Visual communication is the cornerstone of architectural practice. Generative AI image and video models can produce impressive results at first glance, but they may quickly overpower a student’s design intent and blur authorship. Furthermore, local use requires powerful workstations that few students have. Paid online services exist, but they raise issues of privacy, create unequal access, and hinder the development of a shared, collaborative culture, as student work remains siloed.

This project provides institutional infrastructure and equips students with critical thinking and control, ensuring that generative AI serves their creative vision—not the other way around.

What: We deliver a self-paced Moodle course featuring video tutorials and workflow templates. This is paired with easy access to open-source AI tools and a shared gallery of custom “style adapters,” turning individual work into a growing institutional asset.

The educational framework also enables novel research by creating a rich dataset of images paired with their design context, grades, and peer reviews, thereby supporting ongoing work on AI-assisted architectural design assessment.

How: We begin with the mandatory Digital Theory and Culture course (approximately 350 students) and then scale department-wide, aiming for a sustainable model that serves both mandatory courses and electives. We will test two pathways toward a blueprint for scalable, fair, and cost-aware GPU access: on-demand, vetted cloud GPUs and D-ARCH–operated pop-up GPU virtual machines.

• Integrate AI-enhanced visualization into teaching – Embed workflows and models into a structured course, supporting progressive skill-building and iterative design practice.
• Establish blueprint for accessible AI visualization infrastructure – Set up shared GPU VMs for remote access to AI image models workflows via ComfyUI (an open-source AI image workflow environment).
• Create customizable workflows – Develop user-friendly workflows in ComfyUI for generating photorealistic and abstract visualizations from images, 3D models, and text prompts.
• Set up a shared gallery of architecture-specific style adapters – Contains AI image model adapters (LORAs) on curated datasets trained by us, as well as students using the course material.
• Support Collaboration and Research – Document workflows and results to create a curated, shareable skillbuilding template that benefits future teaching, studio work and research at DARCH and ETH.

Students gain future-proof skills at the intersection of architecture and AI, giving them an edge in modern creative industries. The project builds technical, creative, and critical literacy in AI-based visualization, ensures equal access to computing resources and storage, and encourages learning through experimentation. Students leave with reusable assets and portfolios showcasing powerful visualization techniques, as well as a clear stance on authorship, bias, and responsible use.

Lecturers receive ready-to-use, architecture-specific AI workflows and templates. This streamlines course preparation, enriches studio teaching with powerful visualization tools, and allows lecturers to focus their feedback on design quality rather than technical support.

For D-ARCH and ETH, shared workflows and a gallery of style adapters create a sustainable and reproducible foundation for future teaching. The initiative fosters a disciplinary discourse community around image-based AI in education and practice, connects studios, methods, and research, and positions the department as a reference point for innovation in project-based education.