Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

MAGIC: Transition-Aware Generation of Navigable Multi-Scene Game Worlds with Large Language Models

Source

arXiv

arxiv.org/abs/2607.11594v1

Read the source

Source

Plain-Language Brief

The paper presents MAGIC, a system that generates connected multi-scene game worlds using large language models. The system overcomes three challenges in single-scene methods: cross-scene consistency, in-scene navigability, and transition evaluation. It uses a four-stage pipeline to create runnable multi-scene game projects from natural-language prompts. MAGIC also introduces a transition-focused evaluation agent for testing the generated transitions.

What the Document Says

  • The paper proposes MAGIC, a system that generates connected multi-scene game worlds using large language models.
  • MAGIC addresses three obstacles left unsolved by single-scene methods: cross-scene consistency, in-scene navigability, and transition evaluation.
  • The system is a four-stage pipeline turning a natural-language prompt into a runnable multi-scene game project.
  • The pipeline includes planning a shared transition-aware intermediate representation, specifying each scene while enforcing portal reachability with a flood-fill validator, generating scenes and their transition scripts, and combining them into one project.
  • A new benchmark of 100 multi-scene cases was created for testing.
  • MAGIC produces an executable project for every case in the benchmark.
  • On the benchmark, MAGIC reaches 0.99 precision, 0.95 recall, and 0.96 F1 on end-to-end transition identification.
  • The paper also introduces a transition-focused evaluation agent for testing each transition in play.

Key Concepts Mentioned

  • multi-scene game worlds
  • large language models
  • transition-aware
  • prompt-to-project system
  • navigable layouts
  • transition identification
  • evaluation agent

Claims Made by the Document

  • MAGIC generates connected multi-scene game worlds using large language models.
  • MAGIC addresses three obstacles left unsolved by single-scene methods.
  • MAGIC recovers more ground-truth portals and yields markedly more navigable layouts than an LLM baseline and Holodeck.

Evidence Used

  • The paper's content describes the proposed system, its challenges, and its results.

Methods / Approach

  • Four-stage pipeline: planning, specifying scenes, generating scenes and transition scripts, combining into one project.
  • Transition-focused evaluation agent for testing transitions

Limitations Stated by the Authors

Not stated in source

Limitations Noted by the Agent

Agent interpretation — not from the source.

The system's performance may depend on the quality and specificity of the natural-language prompt provided.

Open Questions

  • What is the LLM baseline for comparison?

Notes for Future Q&A

May include interpretation. Labeled where it does.

It would be helpful to know the specifics about the LLM baseline used in the paper for better understanding its comparison with MAGIC.

Raw Abstract / Excerpt

Multi-scene navigation (clearing an objective in one bounded space and then crossing a portal into the next) is a defining feature of contemporary 3D games, but authoring it is laborious: every portal must have consistent endpoints on both sides, each interior must remain navigable once it is furnished, and the resulting connectivity must be kept consistent across many files. Recent large language model (LLM) and multimodal LLM (MLLM) scene generators have made single-interior synthesis dramatically cheaper, yet they produce one scene at a time and cannot, by naive repetition, yield a connected multi-scene world. We identify three obstacles that single-scene methods leave unsolved: cross-scene consistency, in-scene navigability, and the evaluation of whether a transition actually works. We present MAGIC, a prompt-to-project system that addresses all three. MAGIC is a four-stage pipeline that turns a single natural-language prompt into a runnable multi-scene game project: it plans a shared transition-aware intermediate representation, specifies each scene while enforcing portal reachability with a flood-fill validator, generates the scenes together with their transition scripts, and combines them into one project. Because existing single-scene fidelity metrics never execute a transition, we further introduce a transition-focused evaluation agent that runs each transition in play. On a new benchmark of 100 multi-scene cases, MAGIC produces an executable project for every case and reaches 0.99 precision, 0.95 recall, and 0.96 F1 on end-to-end transition identification; stage by stage, it recovers more ground-truth portals and yields markedly more navigable layouts than an LLM baseline and Holodeck. Our code is available at https://github.com/sereneee1201/MAGIC/.