Design2Code introduces the first comprehensive benchmark for evaluating how effectively multimodal language models can transform visual designs directly into functional code implementations.

• Curates 484 diverse real-world webpages as test cases
• Develops automated evaluation metrics to assess code generation quality
• Evaluates performance of leading multimodal AI models on this new front-end engineering task
• Establishes a framework for measuring progress in automated UI implementation

This research represents a significant advancement for engineering teams by potentially automating the labor-intensive process of translating designs into code, streamlining development workflows, and reducing the time from concept to implementation.

Design2Code: Benchmarking Multimodal Code Generation for Automated Front-End Engineering