Lewandofski/OpenVE-3M

---
license: cc-by-nc-4.0
task_categories:
- Video-to-Video
size_categories:
- 1M<n<10M
tags:
- video
---

<h1 align="center" style="line-height: 50px;">
OpenVE-3M: A Large-Scale High-Quality Dataset for Instruction-Guided Video Editing
</h1>

<div align="center">

[Haoyang He^1*](https://scholar.google.com/citations?hl=zh-CN&user=8NfQv1sAAAAJ),
Jie Wang^2*,
[Jiangning Zhang¹](https://zhangzjn.github.io),
[Zhucun Xue¹](https://scholar.google.com/citations?user=m3KDreEAAAAJ&hl=en),
[Xingyuan Bu²](https://scholar.google.com/citations?hl=en&user=cqYaRhUAAAAJ&view_op=list_works),
[Qiangpeng Yang²](https://scholar.google.com/citations?user=vr9z1VQAAAAJ&hl=en&oi=ao),
[Shilei Wen²](https://scholar.google.com/citations?user=zKtYrHYAAAAJ&hl=en&oi=ao),
[Lei Xie^1#](https://scholar.google.com/citations?hl=zh-CN&user=7ZZ_-m0AAAAJ),

¹Zhejiang University, ²Bytedance

\*Equal Contribution. \# Corresponding Author.
</div>

<div align="center">
<a href="https://lewandofskee.github.io/projects/OpenVE/"><img src="https://img.shields.io/static/v1?label=Project%20Page&message=Web&color=green"></a> &ensp;
<a href="https://arxiv.org/abs/2512.07826"><img src="https://img.shields.io/static/v1?label=Tech%20Report&message=Arxiv&color=red"></a> &ensp;
<a href="https://www.modelscope.cn/models/"><img src="https://img.shields.io/static/v1?label=Model&message=ModelScope&color=blue"></a> &ensp;

<a href="https://huggingface.co/Lewandofski/OpenVE-Edit"><img src="https://img.shields.io/static/v1?label=OpenVE-Edit%20Model&message=HuggingFace&color=yellow"></a> &ensp;
<a href="https://huggingface.co/datasets/Lewandofski/OpenVE-Bench"><img src="https://img.shields.io/static/v1?label=OpenVE-Bench&message=HuggingFace&color=yellow"></a> &ensp;
</div>

---

## 🌍 Overview
We introduce OpenVE-3M, an open-source, large-scale, and high-quality dataset for instruction-based video editing. The OpenVE-3M dataset includes eight major video editing categories: six SA (Global Style Transfer,
Background Change, Local Change, Local Remove, Local Add, Subtitles Edit) and two NSA (Multi-Shot
Camera Edit, Creative Edit). SA edits maintain perfect consistency in motion and detail between the original
and edited videos. In contrast, NSA edits maintain the primary subject’s consistency but introduce new,
creative motion.

## Dataset Statistics

- **Total Examples**: `3,000,000+` video editing triplets
- **Video Resolution**: `720P` resolutions (1280\*720 / 720\*1280)
- **Video Length**: `65-129` frames per video
- **Categories**: `8` Categories with Spatially-Aligned Edits (Global Style, Background Change, Local Change, Local Remove, Local Add, and Subtitles Edit)
and Non-Spatially-Aligned Edits (Camera Multi-Shot Edit and Creative Edit)
- **Average Instructions Length**: Average `40.6` words per video

## Dataset Structure

The dataset is organized as follows:

```folder
├── OpenVE-3M
│ ├── videos
| |── global_style.tar.gz.00
| |── ...
| |── background_change..tar.gz.00
| |── ...
| |── local_change..tar.gz.00
| |── ...
| |── local_remove..tar.gz.00
| |── ...
| |── local_add..tar.gz.00
| |── ...
| |── subtitles_edit..tar.gz.00
| |── ...
│ ├── csv_files
| |── global_style.csv
| |── background_change.csv
| |── local_change.csv
| |── local_remove.csv
| |── local_add.csv
| |── subtitles_edit.csv
| |── camera_edit.csv
| |── creative_edit.csv
```

### Dataset Space Usage

- **Global Style (~510G)**
- **Background Change (~170G)**
- **Local Change (~179G)**
- **Local Remove (~453G)**
- **Local Add (~760G)**
- **Subtitles Edit (~164G)**
- **Multi-Shot Camera Edit (~981G)**
- **Creative Edit (~1.6T)**

### Training CSV Files

Each csv file contains triplet items of:
- `video`: Path to the corresponding edited video
- `original_video`: Path to the source video
- `prompt`: Editing instruction

## Downloading and Extracting the Dataset

### Full Dataset Download

```python
from datasets import load_dataset
# Download the entire dataset
dataset = load_dataset("Lewandofski/OpenVE-3M")
```

### Extracting the Video Data
On Linux/macOS or Windows (with Git Bash/WSL):
```bash
# Navigate to the directory containing the split files
cd path/to/your/dataset/part
# For example, to extract the global_style videos:
cat global_style.tar.gz.* | tar -zxv
```

## Citation
If you find OpenVE useful for your research and applications, please cite using this BibTeX:
```
@article{he2025openve-3m,
title={OpenVE-3M: A Large-Scale High-Quality Dataset for Instruction-Guided Video Editing},
author={Haoyang He, Jie Wang, Jiangning Zhang, Zhucun Xue, Xingyuan Bu, Qiangpeng Yang, Shilei Wen, Lei Xie},
journal={arXiv preprint arXiv:2512.07826},
year={2025}
}
```

## LICENSE
OpenVE-3M is licensed under the CC-BY-NC-4.0 License.