WorldAgents: Can Foundation Image Models be Agents for 3D World Models?

TL;DR

This paper investigates whether 2D foundation image models inherently possess 3D world modeling capabilities by evaluating and leveraging them through a multi-agent system for 3D scene synthesis and reconstruction.

Contribution

It introduces an agentic framework combining vision-language models and generators to enable 3D world synthesis from 2D foundation models, demonstrating their implicit 3D understanding.

Findings

2D models contain implicit 3D world knowledge

The proposed multi-agent system produces coherent 3D reconstructions

Synthesized scenes are realistic and view-consistent

Abstract

Given the remarkable ability of 2D foundation image models to generate high-fidelity outputs, we investigate a fundamental question: do 2D foundation image models inherently possess 3D world model capabilities? To answer this, we systematically evaluate multiple state-of-the-art image generation models and Vision-Language Models (VLMs) on the task of 3D world synthesis. To harness and benchmark their potential implicit 3D capability, we propose an agentic framing to facilitate 3D world generation. Our approach employs a multi-agent architecture: a VLM-based director that formulates prompts to guide image synthesis, a generator that synthesizes new image views, and a VLM-backed two-step verifier that evaluates and selectively curates generated frames from both 2D image and 3D reconstruction space. Crucially, we demonstrate that our agentic approach provides coherent and robust 3D reconstruction, producing output scenes that can be explored by rendering novel views. Through extensive experiments across various foundation models, we demonstrate that 2D models do indeed encapsulate a grasp of 3D worlds. By exploiting this understanding, our method successfully synthesizes expansive, realistic, and 3D-consistent worlds.