Reimagining Urban Riverfront Revitalization: Extracting Pasig River’s Common Landscape Characteristics on Instagram for GenAI Prompt Engineering

Abstract

Generative Artificial Intelligence (GenAI), particularly text-to-image technology, is becoming more accessible and prevalent. It has become a tool for reimagining urban landscape scenarios (ULSs), in which landscape elements and forms reflect ideal urban living quality for urban landscape studies. This study examines how a generative model can address emerging urban needs, focusing on the Pasig River in Manila, Philippines—the backbone of the city's urban development. To do so, the study explored the depths of landscape aesthetics using a relevant source of landscape information—the online commons—grounded in Post-media Aesthetics and Social Capital theories. Four methods were employed in this study: data mining, AI object detection, prompt engineering, and ULS generation. Using Instagram, image posts with the #PasigRiver hashtag and geotagged within the City of Manila, from pre- and post-Pasig River Development Project Phase 1, were extracted and screened based on a set of landscape characteristics criteria. VisionAI v2.1-2.5 and ChatGPT-4o mini were used to extract and generate summarized visual descriptions from the images. Using an established scene format prompt, extracted summarized visual prompts, and standardized advance prompts, combined prompts were formed. The Midjourney application via a Discord bot was used to process and test the combined prompts to generate the ULSs. The ULS final sampling revealed a landscape composition of a vibrant urban river scenery at night, a prominent bridge, various river-related activities, and linear parks- figuratively reflecting the common landscape characteristics of urban riverfront revitalization. We argue that the use of GenAI image-to-text, text-to-text, and text-to-image technologies can visualize data from the digital realm, demonstrating that the engineering process captures visual cues and emerging meanings from the study area. Ultimately, the multi-method GenAI processes introduced in the study enable GenAI users to facilitate digital placemaking that visually communicates their shifting landscape preferences for URR developments.