[ICLR 2026] MicroVerse: A Preliminary Exploration Toward a Micro-World Simulation
MicroVerse is the first AI framework designed to simulate microscopic biological processes at organ, cellular, and subcellular levels. Unlike existing video generation models that mostly mimic visual textures, MicroVerse integrates physical and biological constraints to faithfully reproduce the mechanisms of life. Leveraging the MicroSim-10K dataset and evaluated against MicroWorldBench, it sets a new benchmark for microscale AI simulations, enabling applications in biomedical research, education, and interactive visualization.
This work represents a leap from “macro-scale visual imitation” to mechanism-aware micro-world modeling, making AI a tool for understanding life at its most fundamental scales.
- 🌱 True Microscale Simulation: Accurately models organ, cellular, and subcellular dynamics, capturing DNA replication, cell division, and apoptosis with scientific fidelity.
- 📊 Benchmarking Science: Evaluated with MicroWorldBench, the first expert-annotated rubric-based test for microscopic simulations.
- 💾 Expert-Verified Dataset: MicroSim-10K provides 9,601 high-resolution, semantically rich video clips guiding AI toward mechanism-level understanding.
- 🧠 Mechanism over Visuals: Moves beyond realistic textures to faithfully reproduce underlying biological laws and temporal dynamics.
- 🎓 Research & Education Ready: Enables cost-effective, interactive exploration of life’s hidden processes, bridging AI, biology, and learning.
[2026.03.03]🚀 Paper released.[2026.01.26]🎉 MicroVerse has been officially accepted as a Poster at ICLR 2026.
If you find our paper and code useful in your research, please consider giving a star ⭐ and citation 📝:
@article{wang2026microverse,
title={MicroVerse: A Preliminary Exploration Toward a Micro-World Simulation},
author={Wang, Rongsheng and Wu, Minghao and Zhou, Hongru and Yu, Zhihan and Cai, Zhenyang and Chen, Junying and Wang, Benyou},
journal={arXiv preprint arXiv:2603.00585},
year={2026}
}