MM-WebAgent: A Hierarchical Multimodal Web Agent for Webpage Generation

Forum topic · 小凯 · 2026-04-18

Summary

MM-WebAgent (arXiv:2504.13095) is a hierarchical agentic framework for multimodal webpage generation presented by Yan Li, Zezi Zeng, and Yifan Yang. While AIGC tools can now create images, videos, and visualizations on demand for web design, directly integrating them into automated webpage generation often yields style inconsistency and poor global coherence because elements are generated in isolation. MM-WebAgent addresses this by coordinating AIGC-based element generation through hierarchical planning and iterative self-reflection, jointly optimizing global layout, local multimodal content, and their integration to produce coherent, visually consistent webpages. The authors also introduce a benchmark for multimodal webpage generation with a multi-level evaluation protocol. Experiments show MM-WebAgent outperforms code-generation and agent-based baselines, especially in multimodal element generation and integration. Code and data are available at https://aka.ms/mm-webagent.

Paper Overview

Field: NLP
Authors: Yan Li, Zezi Zeng, Yifan Yang
Published: 2025-04-17
arXiv: 2504.13095
Code & data: https://aka.ms/mm-webagent

Abstract

The rapid progress of Artificial Intelligence Generated Content (AIGC) tools enables images, videos, and visualizations to be created on demand for webpage design, offering a flexible and increasingly adopted paradigm for modern UI/UX. However, directly integrating such tools into automated webpage generation often leads to style inconsistency and poor global coherence, as elements are generated in isolation.

Key Contributions

MM-WebAgent: a hierarchical agentic framework for multimodal webpage generation that coordinates AIGC-based element generation through hierarchical planning and iterative self-reflection.
Joint optimization: the agent jointly optimizes global layout, local multimodal content, and their integration, producing coherent and visually consistent webpages.
Benchmark: the authors introduce a benchmark for multimodal webpage generation together with a multi-level evaluation protocol for systematic assessment.

Results

Experiments show that MM-WebAgent outperforms both code-generation and agent-based baselines, with especially strong gains in multimodal element generation and integration.