WebCode2M introduces a large-scale, real-world dataset to address the critical gap in training Multimodal Large Language Models (MLLMs) for automated webpage code generation.

• Creates a dataset of 2 million webpage design-code pairs from real-world websites
• Develops WebCoder, a specialized model outperforming existing MLLMs in converting designs to functional code
• Achieves significant improvements in both code quality and visual fidelity metrics
• Establishes new benchmarks for evaluation in webpage code generation

This research dramatically reduces front-end development workload by enabling AI to transform visual designs directly into implementation-ready code, potentially revolutionizing software engineering workflows.

WebCode2M: A Real-World Dataset for Code Generation from Webpage Designs