LaTCoder: Converting Webpage Design to Code with Layout-as-Thought
Yi Gui, Zhen Li, Zhongyi Zhang, Guohao Wang, Tianpeng Lv, Gaoyang Jiang, Yi Liu, Dongping Chen, Yao Wan, Hongyu Zhang, Wenbin Jiang, Xuanhua Shi, Hai Jin
2025-08-07
Summary
This paper talks about LaTCoder, a system that helps turn webpage designs into code more accurately by breaking the designs into smaller parts and using a thinking process called Chain-of-Thought with large language models.
What's the problem?
The problem is that existing methods for converting webpage designs to code often fail to keep the layout looking exactly the same, making the resulting webpages look different from the original designs.
What's the solution?
The solution was to create LaTCoder, which divides the design into blocks and guides the model to think through each block carefully before generating code. This Chain-of-Thought method helps the model understand the layout better and produce code that preserves the design more faithfully.
Why it matters?
This matters because it makes it easier and faster to create high-quality webpages that look just like the designs, saving time for web developers and designers. It improves how AI models assist in coding tasks by making their output more reliable and closer to what users want.
Abstract
LaTCoder enhances layout preservation in design-to-code tasks by dividing webpage designs into blocks and using Chain-of-Thought reasoning with MLLMs, achieving significant improvements in metrics and human preference.