Unlocking the conversion of Web Screenshots into HTML Code with the WebSight Dataset introduces an innovative dataset catered to empower vision-language models (VLMs) to transform UI screenshots into HTML.

• WebSight comprises 2 million HTML and screenshot pairs, serving as a high-quality dataset for model fine-tuning.
• The fine-tuned foundational VLM on WebSight showcases proficiency in generating functional HTML from webpage screenshots.
• Aimed at boosting no-code solution efficiency, the authors release the dataset open-source to further research.

The ability for VLMs to interpret and recreate web page designs from screenshots could revolutionize web development processes, especially in no-code environments. The open-sourcing of the WebSight dataset is expected to spur significant advancements in this domain.