WebViSE extracts visual skills from the open web for website LLMs
Original title: WebViSE: Scaling Visual Design Intelligence from the Open Web for LLM Website Generation
Its skill DAG beats non-hierarchical retrieval at 39.09% of estimated cost with Qwen3.8-Max; skills and code are released.
WebViSE abstracts open-web website implementations into reusable Visual Skills and a structured library organized as a multi-parent DAG. Skill descriptions and DAG structure evolve jointly using Skill-conditioned Potential Queries so user instructions can match relevant skills. The skills can be injected as plug-and-play guidance or used to synthesize training data for open-source models. The authors report better visual quality and interaction fidelity across web-development benchmarks and human evaluation, and release the skills, DAG, and code.
- Auto-evolved retrieval used 39.09% of estimated cost with Qwen3.8-Max.
- Open-source models can internalize skills through a data-synthesis pipeline.
- Extracted Visual Skills, the Skill DAG, and code are on GitHub.
- Authors report gains in visual quality and interaction fidelity.
From the original article
1 min read at arxiv.orgLarge language models have made remarkable strides in website development, enabling complete and interactive web experiences to be built from natural-language instructions with far less effort. Yet, non-expert users often struggle to articulate sophisticated visual and interaction designs in natural language.
By Yifei Sun · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.