Back to AI Research

AI Research

MAGIC: Transition-Aware Generation of Navigable Mul... | AI Research

Key Takeaways

  • MAGIC: Transition-Aware Generation of Navigable Multi-Scene Game Worlds with Large Language Models Creating 3D games with multiple connected rooms—such as du...
  • 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.
  • Because existing single-scene fidelity metrics never execute a transition, we further introduce a transition-focused evaluation agent that runs each transition in play.
  • Our code is available at this https URL .
Paper AbstractExpand

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 this https URL .

MAGIC: Transition-Aware Generation of Navigable Multi-Scene Game Worlds with Large Language Models
Creating 3D games with multiple connected rooms—such as dungeon crawlers or escape rooms—is a complex task that requires designers to manually ensure that doors, portals, and room layouts align perfectly across different files. While recent AI tools have made it easier to generate single rooms, they struggle to create connected, navigable worlds where a player can move seamlessly from one scene to the next. MAGIC is a new system designed to automate this process, turning a single natural-language prompt into a fully functional, multi-scene game project.

Solving the Multi-Scene Challenge

Existing AI models often fail to create multi-scene worlds because they treat each room as an isolated task. This leads to three primary issues: cross-scene inconsistency, where portals don't match up; poor navigability, where furniture accidentally blocks doorways; and a lack of evaluation, as there are no existing tools to test if a transition between scenes actually works. MAGIC addresses these by using a shared "intermediate representation" that acts as a blueprint for the entire game, ensuring that every portal and connection is tracked and verified before the final game files are built.

A Four-Stage Pipeline

MAGIC operates through a structured, four-stage pipeline: 1. Planning: An LLM analyzes the user's prompt to create a "transition-aware automaton," a graph that maps out every scene and how they connect to one another. 2. Scene Specification: The system defines the layout of each room. It uses a "flood-fill" algorithm to create a 2D grid of the room, checking if the player can physically reach every portal. If furniture blocks a path, the system adjusts the layout until the room is fully navigable. 3. Scene Generation: The system builds the 3D meshes for walls, floors, and objects, and attaches "LevelLoader" scripts to the portals. These scripts handle the actual transition logic, such as loading the next scene when a player walks through a door. 4. Combination: The individual scenes are stitched together into a single, runnable Unity project where all transitions are linked and ready for play.

Testing and Results

To ensure the generated games actually work, the researchers developed a transition-focused evaluation agent. Unlike previous metrics that only look at the quality of a single room, this agent acts like a player: it enters the game, walks to portals, and executes transitions to see if they function as intended. In a benchmark of 100 multi-scene cases, MAGIC successfully produced an executable project for every instance. It achieved high precision and recall scores, demonstrating that it can generate complex, interconnected game environments that are both consistent and navigable, significantly outperforming existing baseline models.

Comments (0)

No comments yet

Be the first to share your thoughts!