WeDesign: Generative AI-Facilitated Community Consultations for Urban Public Space Design.

Résumé (uniquement disponible en anglais)

Community consultations are central to urban planning but often constrained by resources, language, and power imbalances. We study how generative text-to-image models can support more equitable dialogue. Using WeDesign, a custom platform integrating Stable Diffusion XL, we ran a half-day workshop in Montréal with five mixed focus groups (residents, designers, AI specialists) and conducted six interviews with urban-planning professionals. Real-time images stimulated creativity, helped participants exchange ideas, and supported iterative refinement. Participants nevertheless reported limits in depicting reduced-mobility needs, local architectural cues, and bilingual prompting; many outputs felt polished yet de-contextualized. Participants and experts recommended open-source tooling with region-specific in-painting, multilingual support, group voting, and preference sliders, alongside structured facilitation. We discuss how generative AI can broaden participation while requiring safeguards and expectations management, contributing empirical evidence on opportunities and limits of text-to-image AI in participatory urban design.