ABSTRACT Generative artificial intelligence (GenAI) is increasingly positioned as a tool for participatory urban planning, yet its capacity to produce scenarios that are both visually credible and functionally meaningful remains empirically underexplored. This study presents and tests a dual-method evaluation framework integrating three analytical components: pixel-based visual greenness assessment using Hue, Saturation, and Value (HSV) thresholding; rubric-scored functional evaluation of spatial affordances; and an exploratory stakeholder survey. The framework was applied to fifteen AI-generated urban greening scenarios across five sites in the post-industrial young city district of Gdańsk, Poland, produced using DALL-E 3 and Canva Pro Magic AI and grounded in the Essex Design Guide for Landscape and Greenspaces. Findings indicate that vegetation density alone does not reliably predict stakeholder preference. Rather, the relationship between green coverage and user satisfaction proved nonlinear and context-dependent, with scenarios combining moderate greenery and diverse functional programming consistently outperforming those optimised along a single dimension. Stakeholder responses varied significantly across groups: local community members expressed the highest levels of support and showed a positive correlation between GenAI familiarity and endorsement, while architecture and urban planning students were notably more skeptical. These findings suggest that GenAI holds practical value as a decision-support instrument in early-stage participatory design, but only when paired with expert review and quantitative evaluation. The framework offers transferable guidance for planners seeking to integrate generative tools into inclusive, evidence-based urban greening workflows.
Annan et al. (Fri,) studied this question.