World Craft: Agentic Framework to Create Visualizable Worlds via Text

TL;DR

World Craft presents a two-component framework, World Scaffold and World Guild, to democratize the creation of executable, visualizable AI Town environments from natural language. By decoupling semantic grounding from spatial grounding through a four-agent collaboration (Semantic Enrichment, Layout Manager, Critic, and Asset Synthesis) and introducing a reverse-synthesis data pipeline with a rich asset library, the approach achieves superior layout realism, consistency, and intent conveyance compared with commercial code agents and baseline LLMs. A multi-stage training strategy and robust evaluation metrics—including collision-free layout, connectivity, object existence, and visual-semantic alignment—validate the framework’s effectiveness and reliability. These results indicate a scalable path toward accessible AI-generated worlds, with practical implications for gaming, simulation, and research into emergent multi-agent environments.

Abstract

Large Language Models (LLMs) motivate generative agent simulation (e.g., AI Town) to create a ``dynamic world'', holding immense value across entertainment and research. However, for non-experts, especially those without programming skills, it isn't easy to customize a visualizable environment by themselves. In this paper, we introduce World Craft, an agentic world creation framework to create an executable and visualizable AI Town via user textual descriptions. It consists of two main modules, World Scaffold and World Guild. World Scaffold is a structured and concise standardization to develop interactive game scenes, serving as an efficient scaffolding for LLMs to customize an executable AI Town-like environment. World Guild is a multi-agent framework to progressively analyze users' intents from rough descriptions, and synthesizes required structured contents (\eg environment layout and assets) for World Scaffold . Moreover, we construct a high-quality error-correction dataset via reverse engineering to enhance spatial knowledge and improve the stability and controllability of layout generation, while reporting multi-dimensional evaluation metrics for further analysis. Extensive experiments demonstrate that our framework significantly outperforms existing commercial code agents (Cursor and Antigravity) and LLMs (Qwen3 and Gemini-3-Pro). in scene construction and narrative intent conveyance, providing a scalable solution for the democratization of environment creation.

Figures (16)

• Figure 1: An illustration of our motivation and goal.
• Figure 2: Architecture of WorldCraft. It comprises the World Guild for intent analysis and layout generation, and the World Scaffold for automated scene construction.
• Figure 3: Two-stage fine-tuning data construction process. Utilizing Gemini-3-Pro as all the agents, we perform 10 runs for each of scenario descriptions. During the filtering process, approximately 5k invalid samples are discarded, and 1.2k long-tail samples undergo human rectification, resulting in a final dataset of approximately 14k samples.
• Figure 4: Results of fine-grained comparison on performance stability under different input lengths in the test set.
• Figure 5: Dynamic changes in model output quality during multi-round correction processes.