Vision2Web
Runnable front-end and full-stack implementations paired with visual references, interaction specifications, PRDs, and programmatic evaluation.
Overview
Copula Vision2Web is a structured training and evaluation corpus for reconstructing runnable frontend and full-stack products from design intent. Each datapoint is a complete, independently launchable task: the problem input (multi-viewport prototypes, interaction specs or a PRD with supporting assets), a reference implementation built from that input, and a programmatically defined evaluation workflow. The dataset trains and evaluates how well models turn requirements and designs into high-quality, runnable frontend and full-stack products.
Built for
- Multimodal post-training on design-to-code
- Fidelity evals: layout, spacing, type, colour
- Agent training on iterative visual refinement
Featured samples
Museum collection archive
Interactive · Web
A scholarly catalogue for a Dutch and Flemish Golden Age collection, rebuilt from design frames — search, artist filtering, sorting and view controls, with every plate and typeface bundled locally.
Build the dataset your model is missing.
Every product here is built with domain experts and delivered to your training or evaluation spec. Tell us what you need.
Discuss your project