"Zero-Shot Prompting Approaches for LLM-based Graphical User Interface Generation"

Zero-shot prompting to create production-ready GUI prototypes.

This paper introduces zero-shot prompting techniques to automatically generate high-fidelity GUI prototypes from natural language descriptions, reducing the time and resources needed for UI/UX design.

-----

https://arxiv.org/abs/2412.11328

💡 Methods in this Paper:

→ Introduces three novel zero-shot prompting approaches: Retrieval-Augmented GUI Generation (RAGG), Prompt Decomposition (PDGG), and Self-Critique GUI Generation (SCGG)

→ RAGG combines GUI retrieval from large repositories with LLM reasoning to guide generation

→ PDGG breaks down the task into smaller sub-tasks like feature extraction and layout structure

→ SCGG implements an iterative improvement loop where the LLM critiques its own generated prototypes

-----

🔍 Key Insights:

→ LLM-based re-ranking significantly outperforms previous approaches with 81.8% average precision

→ SCGG consistently produces better results than baseline approaches

→ Increasing example count in RAGG (k=7) improves overall GUI quality

→ Content generation enhances visual appeal and user satisfaction

-----

📊 Results:

→ Over 3,000 GUI annotations from 100+ UI/UX experts validate the approaches

→ SCGG outperforms baselines across most metrics

→ RAGG with 7 examples shows significant improvement in feature implementation

→ LLM-based content generation improves visual appeal by 85%