---
license: cc-by-nc-4.0
task_categories:
- image-to-video
---

# OpenVE-3M

This repository contains the large-scale, high-quality dataset for instruction-guided video editing presented in the paper [OpenVE-3M: A Large-Scale High-Quality Dataset for Instruction-Guided Video Editing](https://huggingface.co/papers/2512.07826).

**Project page:** [https://lewandofskee.github.io/projects/OpenVE](https://lewandofskee.github.io/projects/OpenVE)  
**Code:** [https://github.com/OpenVE-Team/OpenVE-3M](https://github.com/OpenVE-Team/OpenVE-3M)

The dataset comprises two primary categories of edits:
- **Spatially-aligned edits:** Global Style, Background Change, Local Change, Local Remove, Local Add, and Subtitles Edit
- **Non-spatially-aligned edits:** Camera Multi-Shot Edit and Creative Edit

All edit types are generated via a meticulously designed data pipeline with rigorous quality filtering. The dataset is intended for training and evaluating instruction-based video editing models.