InfiniDream: Open-Ended Photorealistic 3D Worlds Through Procedural Generation (arXiv 2610.26853)

Forum topic · 小凯 · 2026-10-21

Summary

InfiniDream is a VLM-driven framework for open-ended procedural generation of photorealistic 3D worlds, presented in arXiv paper 2610.26853 by Yi-Chuan Huang, Hao-Jen Chien, Zheng-Hui Huang, Yu-Lun Liu, Kaipeng Zhang, and Zhixiang Wang. Procedural generation offers greater controllability, editability, and diversity than static meshes, but VLM-based code generation typically yields simple assets while photorealistic systems demand heavy manual development. InfiniDream addresses this with two contributions: Symbolically Constrained Asset Generation, which leverages existing procedural algorithms to create novel photorealistic assets while preserving editability and seed-based diversity; and an end-to-end text-to-world pipeline combining semantic scene planning with deterministic frustum-guided grid placement to build large-scale 3D worlds with precise object placement and animation. Evaluations on 100 object and 20 scene generation tasks show improvements across four perceptual metrics, with diversity analysis, ablation studies, and human preference evaluations further validating the framework.

Paper Overview

Research Area: Computer Vision Authors: Yi-Chuan Huang, Hao-Jen Chien, Zheng-Hui Huang, Yu-Lun Liu, Kaipeng Zhang, Zhixiang Wang Published: 2026-10-19 arXiv: 2610.26853

Abstract

Procedural generation plays a key role in synthetic data creation and game development, offering greater controllability, editability, and diversity than static mesh representations. However, VLM-based code generation often produces relatively simple 3D assets, while sophisticated photorealistic procedural systems require extensive manual development. Existing VLM-guided procedural methods remain largely constrained by predefined factories and APIs, limiting their ability to create novel asset types.

We introduce InfiniDream, a VLM-driven framework for open-ended procedural 3D world generation. First, we propose Symbolically Constrained Asset Generation, which leverages existing procedural algorithms to create novel, photorealistic assets while preserving editability and seed-based diversity. Second, we develop an end-to-end text-to-world pipeline that combines semantic scene planning with deterministic frustum-guided grid placement to construct large-scale 3D worlds with precise object placement and animation.

Results

Evaluations on 100 object and 20 scene generation tasks show improvements across four perceptual metrics. Diversity analysis, ablation studies, and human preference evaluations further validate the effectiveness of the framework.

---

*Auto-collected on 2026-10-21*