# MotionDreamer: Zero-Shot 3D Mesh Animation from Video Diffusion Models

**Authors:** Lukas Uzolas, Elmar Eisemann, Petr Kellnhofer  
**arXiv:** [2405.20155](https://arxiv.org/abs/2405.20155)  
**Date:** 2024-05-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2405.20155v1/

## Key points

- Authors Lukas Uzolas, Elmar Eisemann, and Petr Kellnhofer introduce a novel technique for automatically animating arbitrary 3D shapes
- The method leverages a motion prior extracted from a video diffusion model to bring digital 3D worlds and characters to life
- Utilizes an explicit mesh-based representation compatible with existing computer graphics pipelines for seamless integration
- Enhances motion fitting accuracy by leveraging diffusion features in the approach
- Experimentally validated using two different diffusion models and four animation models, demonstrating superior performance in re-animating diverse 3D shapes
- Confirmed through a user study conducted by the authors that the time-efficient zero-shot method outperforms existing techniques
- "MotionDreamer" presents an innovative solution for automatically animating 3D shapes based on motion priors extracted from video diffusion models.

## Summary

In their paper titled "MotionDreamer: Zero-Shot 3D Mesh Animation from Video Diffusion Models," authors Lukas Uzolas, Elmar Eisemann, and Petr Kellnhofer introduce a novel technique for automatically animating arbitrary 3D shapes. The proposed method leverages a motion prior extracted from a video diffusion model to bring digital 3D worlds and characters to life. By utilizing an explicit mesh-based representation that is compatible with existing computer graphics pipelines, the authors ensure seamless integration into current workflows. Additionally, they enhance the accuracy of motion fitting by leveraging diffusion features in their approach. The efficacy of these features for animation fitting is thoroughly analyzed, and the authors experimentally validate their method using two different diffusion models and four animation models. Through these experiments, they demonstrate that their time-efficient zero-shot method outperforms existing techniques in re-animating a diverse set of 3D shapes. This superior performance is further confirmed through a user study conducted by the authors. Overall, "MotionDreamer" presents an innovative solution for automatically animating 3D shapes based on motion priors extracted from video diffusion models. The project website provides additional information and resources for those interested in exploring this cutting-edge technology further.