Plan, then construct
The Orchestrator retrieves skills. The Planner organizes semantic subtasks. Fresh Builders query the persistent scene state and carry out each task.
Portable game scene generation
A Multi-Agent Framework for
Portable Game Scene Generation
1 Tencent2 Fudan University
Construct rich game scenes with coordinated agents.
Reuse the process and experience across engines.
The idea
Moving a scene generator to a new engine requires more than exporting its output. Its planning, tools, and stored experience may depend on the original backend.
SceneAgent separates persistent scene state, task-local reasoning, and reusable skills. Coordinated agents construct the scene; backend adapters handle how it is realized in each engine.
Scene collection
From city streets to imagined worlds.
Selected frames from the project video.
Built to travel
The same construction protocol and skill library,
retargeted through backend integration components.
Reused Agent protocol, SceneContext, core tools, and semantic skills.
Adapted Backend adapter, asset catalog and database, and capability-tool wrappers.
Qualitative examples of retargeting. The engines use different asset catalogs; these are not matched quality comparisons.
Inside SceneAgent
Separate what the world remembers
from what each agent needs to reason about.
The Orchestrator retrieves skills. The Planner organizes semantic subtasks. Fresh Builders query the persistent scene state and carry out each task.
A canonical SceneContext stores the evolving world. Adapters handle coordinate conventions, asset bindings, and export to the target engine.
Human-approved traces become asset-ID-free Layout skills and semantic Content skills, ready to retrieve and recombine in later generations.
GameSceneBench
69 prompts: 56 outdoor and 13 indoor.
Common assets and evaluation protocol in Engine A.
Tie-adjusted preference across
1,860 human judgments.
| Comparator | Human preference |
|---|---|
| AutoUE† | 89.4% |
| Holodeck | 85.2% |
| ImperativeScene | 78.2% |
| Reason-3D | 81.2% |
| SceneWeaver | 82.7% |
50% indicates parity. Ties count as half a preference.
† AutoUE uses its original scene-layout stage adapted to the common backend.
The clearest ablation gains are in prompt alignment and scene completeness, with additional runtime. Spatial plausibility remains a challenge. See the paper for full comparisons, protocols, and limitations.
Citation
@article{li2026sceneagent,
title={SceneAgent: A Multi-Agent Framework for Portable Game Scene Generation},
author={Li, Xingyi and Yang, Jiahao and Zhu, Yiqin and Ye, Jingwen and Sun, Yang and Zhang, Weidong},
journal={arXiv preprint arXiv:2611.17821},
year={2026}
}