# Sketch2Code: Evaluating Vision-Language Models for Interactive Web Design Prototyping > Sketch2Code benchmarks vision-language models on turning rough hand-drawn wireframe sketches into HTML webpage prototypes. Crowd annotators drew 731 sketches of 484 real-world webpages from the Design2Code test set. Ten models, eight commercial and two 8B open-source, generate pages directly and in two multi-turn settings with a GPT-4o simulated user: following its feedback, or asking it clarification questions. Outputs are scored by Design2Code visual similarity and a new IoU-based layout similarity over component types such as text blocks, images, navigation bars, forms and buttons. Claude 3.5 Sonnet and GPT-4o lead single-turn layout similarity, and feedback rounds help commercial models more than question asking does. - Authors: Ryan Li; Yanzhe Zhang; Diyi Yang - Venue: NAACL 2025 - Status: Conference paper - Publication / posting date: April 2025 - Date source: The ACL Anthology record gives the proceedings month as April 2025; Crossref records only the year. The conference ran from 29 April to 4 May 2025, and no publication day is inferred. - First arXiv submission: October 21, 2024 - arXiv: 2410.16232 - Reviewed arXiv version: v1 - DOI: 10.18653/v1/2025.naacl-long.198 - Sources reviewed: 2026-10-27 - Added to the catalog: 2026-11-03 - Original paper: - PDF: - Code: - Project: - BibTeX: - Catalog page: - Record JSON: - Chinese version: [简体中文](https://awesomerea.org/zh/papers/paper-sketch2code/index.md) ## Scope and limitations The generated pages are prototypes scored by visual and layout similarity to reference screenshots, not recovered source code; behavior, interactivity and code equivalence are not evaluated, and page images are replaced with a placeholder. Sketches deliberately omit most text and styling, so similarity scores are low in absolute terms, with best single-turn layout similarity around 22%. The simulated user is GPT-4o, judged faithful in 83.3% of feedback and accurate in 86.7% of answers by human annotators. Only 8B open-source models were run, and a multi-turn example needs 40,000 to 160,000 input tokens. The user study interviewed eight UI/UX experts. Results describe 2024 model versions. ## Why it fits this collection Adds sketch-based and multi-turn interactive webpage generation, complementing the screenshot- and design-based website reconstruction benchmarks in this section. ## Publication and review notes The ACL Anthology page, final PDF and Crossref establish the three-author order, DOI, NAACL 2025 Volume 1 proceedings and pages 3921–3955. Reviewed arXiv v1 (2024-10-21), the only version, commented “preprint, 9 pages”; the description follows the 35-page proceedings PDF. The paper links its code and data through the project page. At the reviewed 21dfd1d commit, SALT-NLP/Sketch2Code is MIT-licensed, pins Python packages including torch 2.4.0, transformers 4.44.1 and playwright 1.43.0, and requires OpenAI, Anthropic or Gemini API keys for the commercial models. The Hugging Face dataset card declares ODC-By, and the paper says the source webpages were collected by Design2Code under ODC-By. ## Primary sources 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. ## About this entry Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX. Original editorial content: MIT. Linked projects, papers and assets retain their own rights and licenses.