Generative AI text-to-image for community participation in landscape planning

Abstract

Effective landscape planning relies on community insights through participatory design to achieve local needs. Visual media can assist community engagement, and visuals created using generative AI text-to-image models are increasingly adopted for such purposes. We explore a new approach of including generative images in participatory planning through a case study with the Diverse Corn Belt Project in the US Corn Belt. Our method is applicable to other contexts, and adds to the literature in three ways. First, we propose a compromise between real-time image generation and extended time workflows of translating participatory discussions into generative images, benefiting from the instant generation of generative models while controlling the output. Building on this proposed pace, we suggest creating what we call ‘controlled imperfect’ images as a balance between “fake perfects” and “conversational imperfects” suggested by the literature. In addition, we propose simplifying the process of translating participatory discussions into an image output through directly collecting keywords necessary for prompt engineering. We build on our case study to outline a revised method for future research.