InfiniteWeb: Scalable Web Environment Synthesis for GUI Agent Training

TL;DR

InfiniteWeb presents a scalable framework to synthesize functional web environments for GUI agent training by aligning tasks, data models, and interfaces through a Unified Specification Stage, validating task-relevant correctness with Task-Centric Test-Driven Development, and guiding frontend design with design-guided generation. It couples automatic evaluators that provide dense reward signals to reinforcement learning, enabling more efficient learning and richer task signals. Empirical results show InfiniteWeb surpassing baselines in functional correctness, achieves high visual fidelity to design references, and meaningfully improves agent performance on OSWorld and Online-Mind2Web, demonstrating practical impact for cross-domain GUI automation. The work addresses core limitations of prior benchmarks—scale, diversity, and verifiable evaluation—facilitating broader, more realistic GUI agent training and benchmarking at scale.

Abstract

GUI agents that interact with graphical interfaces on behalf of users represent a promising direction for practical AI assistants. However, training such agents is hindered by the scarcity of suitable environments. We present InfiniteWeb, a system that automatically generates functional web environments at scale for GUI agent training. While LLMs perform well on generating a single webpage, building a realistic and functional website with many interconnected pages faces challenges. We address these challenges through unified specification, task-centric test-driven development, and a combination of website seed with reference design image to ensure diversity. Our system also generates verifiable task evaluators enabling dense reward signals for reinforcement learning. Experiments show that InfiniteWeb surpasses commercial coding agents at realistic website construction, and GUI agents trained on our generated environments achieve significant performance improvements on OSWorld and Online-Mind2Web, demonstrating the effectiveness of proposed system.

Figures (31)

• Figure 1: Overview of InfiniteWeb. Given a website seed and design image, our system produces a functional website with tasks and evaluators through four stages: the Unified Specification Stage generates tasks and derives data models and interfaces; the Task-Centric Backend and Design-Guided Frontend execute in parallel; and the Evaluator Generation creates task-specific evaluators for dense reward signals.
• Figure 2: Unified Specification Stage. Given a website seed and design image, this stage generates realistic tasks, then derives shared interface design consisting of data models and programming interfaces across pages.
• Figure 3: Task-Centric Backend and Design-Guided Frontend in parallel. The backend uses TCTDD to iteratively generate and validate business logic. The frontend extracts visual styles and generates pages.
• Figure 4: LLM-as-Judge visual quality evaluation. Each pair shows win rates for ours (left) vs baseline (right).
• Figure 5: GUI agent performance improves with more training data generated by InfiniteWeb. Dashed lines indicate potential for further scaling.