Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MAxPrototyper: A Multi-Agent Generation System for Interactive User Interface Prototyping

Mingyue Yuan, Jieshan Chen, Aaron Quigley

TL;DR

MAxPrototyper addresses the challenge of producing editable, coherent, high-fidelity UI prototypes by deploying a multi-agent design system centered on a Theme Design Agent. The framework integrates knowledge-base construction (UI and Icon KBs), knowledge retrieval via embedding-based similarity, theme description and image generation, and specialized content agents for text, images, and icons, all coordinated through a cache-enabled control flow. Empirical results show the approach outperforms baselines in both quality (lower FID) and diversity (higher GD), with ablation studies highlighting the critical roles of knowledge retrieval and theme-image generation. The work advances interactive UI prototyping by delivering theme-consistent, customizable prototypes exportable as SVG or JSON and informs future enhancements to wireframe automation, dynamic components, and design-tool integration.

Abstract

In automated user interactive design, designers face key challenges, including accurate representation of user intent, crafting high-quality components, and ensuring both aesthetic and semantic consistency. Addressing these challenges, we introduce MAxPrototyper, our human-centered, multi-agent system for interactive design generation. The core of MAxPrototyper is a theme design agent. It coordinates with specialized sub-agents, each responsible for generating specific parts of the design. Through an intuitive online interface, users can control the design process by providing text descriptions and layout. Enhanced by improved language and image generation models, MAxPrototyper generates each component with careful detail and contextual understanding. Its multi-agent architecture enables a multi-round interaction capability between the system and users, facilitating precise and customized design adjustments throughout the creation process.

MAxPrototyper: A Multi-Agent Generation System for Interactive User Interface Prototyping

TL;DR

MAxPrototyper addresses the challenge of producing editable, coherent, high-fidelity UI prototypes by deploying a multi-agent design system centered on a Theme Design Agent. The framework integrates knowledge-base construction (UI and Icon KBs), knowledge retrieval via embedding-based similarity, theme description and image generation, and specialized content agents for text, images, and icons, all coordinated through a cache-enabled control flow. Empirical results show the approach outperforms baselines in both quality (lower FID) and diversity (higher GD), with ablation studies highlighting the critical roles of knowledge retrieval and theme-image generation. The work advances interactive UI prototyping by delivering theme-consistent, customizable prototypes exportable as SVG or JSON and informs future enhancements to wireframe automation, dynamic components, and design-tool integration.

Abstract

In automated user interactive design, designers face key challenges, including accurate representation of user intent, crafting high-quality components, and ensuring both aesthetic and semantic consistency. Addressing these challenges, we introduce MAxPrototyper, our human-centered, multi-agent system for interactive design generation. The core of MAxPrototyper is a theme design agent. It coordinates with specialized sub-agents, each responsible for generating specific parts of the design. Through an intuitive online interface, users can control the design process by providing text descriptions and layout. Enhanced by improved language and image generation models, MAxPrototyper generates each component with careful detail and contextual understanding. Its multi-agent architecture enables a multi-round interaction capability between the system and users, facilitating precise and customized design adjustments throughout the creation process.
Paper Structure (19 sections, 2 equations, 1 figure, 3 tables)

This paper contains 19 sections, 2 equations, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Knowledge Base Construction
  4. UI Knowledge Base
  5. Icon Knowledge Base
  6. Theme Design Agent
  7. Knowledge Retrieval
  8. Theme Description Generation
  9. Theme Image Generation
  10. Sub-agent Execution
  11. Text Content Agent
  12. Image Content Agent
  13. Icon Agent
  14. Experiment
  15. Experiments Setup
  16. ...and 4 more sections

Figures (1)

  • Figure 1: Overview: using user prompt and UI Layout as input, MAxPrototyper utilizes a multi-agent method for UI prototype creation. It integrates four primary agents—Theme Design, Text Content, Image Content, and Icon—with Theme Design as the central guiding the sub-agents.