This repo contains the trained model weights and evaluation dataset introduced in

Beyond Pixel Histories: World Models with Persistent 3D State

Samuel Garcin, Thomas Walker, Steven McDonagh, Tim Pearce, Hakan Bilen, Tianyu He, Kaixin Wang, Jiang Bian

Introduction

TL;DR PERSIST (ICML 2026) is a world model that generates coherent rollouts over thousands of steps by modelling a dynamic 3D world state.

✨ Core Capabilities:

• 🖼️ Generates diverse and coherent 3D environments from a single image
• 🛠️ Dynamic World Editing: Supports 3D edits mid-generation
• 🌊 Models unseen interactions & persistent world dynamics

🚀 Quick Start

Please refer to our Github repo

📖 Citation

If you find this project helpful, please consider citing:

@inproceedings{garcin2026beyond,
  title={Beyond Pixel Histories: World Models with Persistent 3D State},
  author={Garcin, Samuel and Walker, Thomas and McDonagh, Steven and Pearce, Tim and Bilen, Hakan and He, Tianyu and Wang, Kaixin and Bian, Jiang},
  booktitle={Forty-third International Conference on Machine Learning},
  year={2026}
}