FracGen: Fracture-Aware Video Generation with Physics-Informed Learning (arXiv 2609.38152)

Forum topic · 小凯 · 2026-10-01

Summary

FracGen is a fracture-aware video generation model that produces plausible, controllable breaking dynamics from a single image of an intact object, conditioned on physics signals. It is built on FracSim, a simulation framework that augments material point method (MPM) simulations with a continuum damage model, generating paired fracture videos and dense, pixel-aligned physical fields at no extra cost beyond standard rendering. FracGen jointly predicts physical fields alongside RGB video and is trained with physics-informed losses that enforce consistency among predicted maps, encouraging the model to capture physical state rather than surface appearance. Without costly test-time simulation or per-scene tuning, FracGen captures material-specific fracture behavior and offers fine-grained control over where objects tear, crack propagation speed, and pre-fracture deformation. The authors also introduce a benchmark for evaluating physical plausibility of generated fracture videos, showing that FracGen outperforms existing video generation baselines in both physical and visual fidelity. Paper: arXiv 2609.38152, by Trong-Tung Nguyen, Jiahan Zhang, and Anand Bhattad (posted 2026-09-29).

Overview

Field: Computer Vision Authors: Trong-Tung Nguyen, Jiahan Zhang, Anand Bhattad Posted: 2026-09-29 arXiv: 2609.38152

Introduction

We introduce FracGen, a fracture-aware video generation model that produces plausible, controllable fracture dynamics from a single image of an intact object, conditioned on physics signals.

To train FracGen, we build FracSim, a fracture-aware simulation framework that augments material point method (MPM) simulation with a continuum damage model, producing paired fracture videos and dense, pixel-aligned physical fields at no additional cost beyond standard rendering.

Method

FracGen leverages these maps in two ways:

It is trained to jointly predict the physical fields alongside RGB video, encouraging the model to capture physical state rather than surface appearance.
It is supervised with physics-informed losses that encourage consistency among the predicted maps.

Results

As a result, FracGen captures distinct material-specific fracture behavior without expensive test-time simulation or per-scene tuning, while providing fine-grained control over:

where an object tears,
crack propagation speed,
and the amount of deformation before fracture.

We further introduce a benchmark for evaluating the physical plausibility of generated fracture videos, and show through extensive experiments that FracGen outperforms existing video generation baselines in both physical and visual fidelity.

Results are best viewed on the project website.

---

*Auto-collected on 2026-10-01.*