WebSight dataset enables screenshot-to-HTML conversion; Sightseer demonstrates Tailwind-based HTML generation

AI Impact Summary

WebSight-v0.2 expands to 2 million screenshot–HTML pairs, incorporating real images and Tailwind CSS to train vision-language models for converting UI designs into functional HTML. Sightseer, fine-tuned on WebSight, demonstrates high-fidelity HTML generation that can embed images from the source screenshots, indicating stronger fidelity in code output. Open-sourcing WebSight lowers the barrier for building low-code UI tooling, potentially accelerating front-end delivery while embedding Tailwind-centric patterns in generated code.