In automated user interface (UI) design generation, a key challenge is the lack of support for iterative processes, as most systems only focus on end-to-end generation of designs as starting points. This results from (1) limited capabilities to fully interpret user design intent from text or images, and (2) a lack of transparency, which prevents designers from refining intermediate results. To address existing limitations, we introduce PrototypeAgent, a human-centered, multi-agent system for automated UI generation. The core of PrototypeAgent is a theme design agent that clarifies implicit design intent through prompt augmentation, coordinating with specialized sub-agents to generate specific components. Designers interact with the system via an intuitive interface, providing natural language descriptions and layout preferences. During generation, PrototypeAgent enables designers to refine generated intermediate guidance or specific components, ensuring alignment with their intent throughout the generation workflow. Evaluations through experiments and user studies show PrototypeAgent's effectiveness in producing high-fidelity prototypes that accurately reflect design intent as well as its superiority over baseline models in terms of both quality and diversity.
View on arXiv