EasyOmnimatte: Taming Pretrained Inpainting Diffusion Models for End-to-End Video Layered Decomposition

Yihan Hu¹, Xuelin Chen² and Xiaodong Cun^{1, 📮}

¹GVC Lab, Great Bay University, ²Adobe Research

TL;DR: EasyOmnimatte splits videos into layers (with effects) using 2-step diffusion without any post optimization.

---

✅ TODO List

We are working on organizing the code. The following items will be released:

• Release ArXiv paper.
• Release Project Page.
• Release inference code (Demo).
• Release pretrained models (Checkpoints).
• Release training scripts and data preparation guidelines.

---

🚀 Getting Started (Coming Soon)

The code is currently being organized and will be released upon acceptance. Please stay tuned!

📧 Contact

If you have any questions, please feel free to email [18281128hyh@gmail.com].