GardenDesigner: Encoding Aesthetic Principles into Jiangnan Garden Construction via a Chain of Agents

2026-04-02

AI Summary

This work addresses the challenges of manual modeling of Jiangnan gardens, which heavily relies on expert knowledge, is time-consuming, and difficult to scale. To overcome these limitations, the authors propose GardenDesigner, a novel framework that, for the first time, systematically encodes the aesthetic principles of Jiangnan gardens and integrates them with a multi-agent collaboration mechanism and an expert knowledge base named GardenVerse to enable automated, culturally constrained procedural modeling. The framework allows non-expert users to interactively generate aesthetically compliant garden layouts in Unity through natural language instructions, covering the full design pipeline—including terrain, pathways, and asset placement. Experimental results demonstrate that the system can produce diverse, high-quality, and stylistically valid garden designs within one minute, with human evaluations confirming its effectiveness and usability.

Abstract

Jiangnan gardens, a prominent style of Chinese classical gardens, hold great potential as digital assets for film and game production and digital tourism. However, manual modeling of Jiangnan gardens heavily relies on expert experience for layout design and asset creation, making the process time-consuming. To address this gap, we propose GardenDesigner, a novel framework that encodes aesthetic principles for Jiangnan garden construction and integrates a chain of agents based on procedural modeling. The water-centric terrain and explorative pathway rules are applied by terrain distribution and road generation agents. Selection and spatial layout of garden assets follow the aesthetic and cultural constraints. Consequently, we propose asset selection and layout optimization agents to select and arrange objects for each area in the garden. Additionally, we introduce GardenVerse for Jiangnan garden construction, including expert-annotated garden knowledge to enhance the asset arrangement process. To enable interaction and editing, we develop an interactive interface and tools in Unity, in which non-expert users can construct Jiangnan gardens via text input within one minute. Experiments and human evaluations demonstrate that GardenDesigner can generate diverse and aesthetically pleasing Jiangnan gardens. Project page is available at https://monad-cube.github.io/GardenDesigner.

Problem

• Jiangnan gardens
• procedural modeling
• aesthetic principles
• digital asset generation
• layout design

Innovation

• procedural modeling
• multi-agent system
• aesthetic encoding
• Jiangnan garden generation
• interactive design