Event

InfiniDream: A VLM-Driven Framework for Open-Ended Procedural 3D World Generation

1 reports · 1 sources

The InfiniDream paper author team proposes InfiniDream, an open-ended procedural 3D world generation framework driven by vision-language models (VLMs), comprising two core methods: symbolically constrained asset generation and a text-to-world pipeline. The authors state that evaluations on 100 object generation tasks and 20 scene generation tasks show improvements across four perceptual metrics. Symbolically constrained asset generation uses existing procedural algorithms to generate realistic assets while preserving editability and seed-based diversity; the text-to-world pipeline combines semantic scene planning with deterministic frustum-guided mesh placement to build large-scale 3D worlds, supporting precise object placement, procedural filling, and animation. The paper was selected for HuggingFace Daily Papers on October 19, 2026.