InfiniDream: A VLM-driven open-ended procedural 3D world generation framework
What happened
The InfiniDream paper (by Yi-Chuan Huang et al.) proposes an open-ended procedural 3D world generation framework driven by a vision-language model (VLM) for generating novel photorealistic assets. The authors report improvements on four perceptual metrics across 100 object and 20 scene generation tasks, supported by diversity analysis, ablation studies, and human preference evaluation. These are self-reported results that have not been independently verified. The paper presents two methods: symbolically constrained asset generation, and deterministic frustum-guided grid placement, which together support a large-scale text-to-world pipeline with object placement, procedural crowds, and animation. Two reports describe the framework's reliance on asset generation inconsistently. An earlier report said it generates assets "on top of existing procedural algorithms," while a later one said it "does not depend on preset factories and APIs" and added a claim about semantic scene planning. Both descriptions are paraphrases of the paper's abstract, so the original paper should be checked to determine which is accurate.
AI-written from the coverage · updated 2026-10-21
Coverage timeline
Follow the coverage to see the story from different sides.
- arXiv Game AIInfiniDream: Building open-ended photorealistic 3D worlds through procedural generation
InfiniDream proposes an open-ended, VLM-driven framework for procedural 3D world generation that produces novel photorealistic assets on top of existing procedural algorithms. It introduces symbolically constrained asset generation and deterministic frustum-guided mesh placement for a large-scale text-to-world pipeline with object placement, procedural crowds, and animation. On tasks generating 100 objects and 20 scenes, the authors report gains on four perceptual metrics, supported by diversity analysis, ablation studies, and human preference evaluations.
- arXiv Game AIInfiniDream: Open-ended photorealistic 3D worlds through procedural generation
InfiniDream proposes a VLM-driven framework for open-ended procedural 3D world generation that creates novel photorealistic assets without relying on preset factories and APIs. Evaluated on 100 object and 20 scene generation tasks, it shows improvements on four perceptual metrics. The framework also combines semantic scene planning with deterministic frustum-guided grid placement to build large-scale 3D worlds with procedural population and animation.