File size: 5,150 Bytes
8f1d68a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
---
pretty_name: EffectData
license: apache-2.0
task_categories:
- image-to-video
- text-to-video
language:
- en
- zh
size_categories:
- 100K<n<1M
configs:
  - config_name: example_preview
    data_dir: example_preview

---

# EffectData

This repository contains the dataset released with the paper **"EffectMaker: Unifying Reasoning and Generation for Customized Visual Effect Creation"**.
> [Shiyuan Yang](https://scholar.google.com/citations?user=gIVxcEoAAAAJ&hl)1,2,†,*, [Ruihuang Li](https://scholar.google.com/citations?user=8CfyOtQAAAAJ&hl)1,†, [Jiale Tao](https://scholar.google.com/citations?user=WF5DPWkAAAAJ&hl=en)1, [Shuai Shao](https://www.sshao.com/)1,‡, [Qinglin Lu](https://openreview.net/profile?id=%7EQinglin_Lu2)1,✉, [Jing Liao](https://www.cityu.edu.hk/stfprofile/jingliao.htm)2,✉  
> 1Tencent Hunyuan 2City University of Hong Kong †Equal Contribution *Work Done During Internship at Tencent Hunyuan ‡Project Lead ✉Corresponding Authors

     

We introduce EffectData, the largest and high-quality synthetic dataset to date, containing 130k videos across 3k diverse effect categories, including atmospheric, transformation, stylistic, motion-based effects, and more. EffectData expands the effect category by an order of magnitude compared with existing datasets and offers English-Chinese bilingual annotations. 

![](assets/cover.png)




## 📁 Directory Structure

```text
EffectData/
├── Videos/                         # Full-resolution MP4 videos, packaged as one zip per effect class.
│   └── <effect_name_en>.zip        # One zip archive per effect class.
├── example_preview/                # Low-resolution preview videos, one example per effect class.
│   └── <effect_name_en>,<effect_name_zh>.mp4 
├── annotations.json                # EN-ZH annotations for all videos.
├── effect_names_list.csv           # List of all effect classes with EN-ZH names.
└── README.md                      
```

- `Videos/` — Full-resolution MP4 data with the short side at 704 pixels. Videos are organized by visual effect class, each provided as a `.zip` archive:
  - Path format: `Videos/<effect_name_en>.zip`
  - After extracting, the videos inside correspond to that visual effect class.
- `example_preview/` — One low-resolution example MP4 per visual effect class. File names include both English and Chinese effect names for convenient browsing before downloading the full archives.
  - Path format: `example_preview/<effect_name_en>,<effect_class_name_zh>.mp4`
- `effect_names_list.csv` — Lists all visual effect classes with bilingual names.
- `annotations.json` — Annotations for all videos. Each entry is keyed by video file name and includes:
  - `video_path` — relative path to the video
  - `prompt_zh` / `prompt_en` — bilingual video prompts
  - `vfx_zh` / `vfx_en` — visual effect labels
  - `instruction_zh` / `instruction_en` — effect instructions
  - `abstract_zh` / `abstract_en` — abstract effect names

  Example entry:
  ```json
  {
    "<video_file_name>.mp4": {
      "video_path": "<effect_name_en>/<video_file_name>.mp4",
      "prompt_zh": "...",
      "prompt_en": "...",
      "vfx_zh": "...",
      "vfx_en": "...",
      "instruction_en": "...",
      "instruction_zh": "...",
      "abstract_zh": "...",
      "abstract_en": "..."
    }
  }
  ```

## 👁️ Preview

You can quickly browse low-resolution example videos for each visual effect class directly in the **Dataset Viewer** at the top of this page (or click **Data Studio** tab).

In the Dataset Viewer's search bar, type either the English or the Chinese name of an effect (refer to `effect_names_list.csv` for full name list) to filter the table and locate the corresponding preview MP4 (which is stored under `example_preview/`). This is the fastest way to inspect what an effect looks like before downloading.

## ⬇️ Download

#### Download the Full Dataset

You can download the complete dataset (~820 GB) with the Hugging Face CLI:

```bash
hf download ysy31415926/EffectData \
  --repo-type dataset \
  --local-dir /your/local/path
```

#### Download One Effect Class

If the effect class name is `<effect_name_en>`, you can directly download the corresponding zip archive with `wget`:

```bash
EFFECT_NAME="<effect_name_en>"
wget -O "${EFFECT_NAME}.zip" \
  "https://huggingface.co/datasets/ysy31415926/EffectData/resolve/main/Videos/${EFFECT_NAME}.zip?download=true"
```

For example, to download `Butterfly_wings.zip`:

```bash
EFFECT_NAME="Butterfly_wings"
wget -O "${EFFECT_NAME}.zip" \
  "https://huggingface.co/datasets/ysy31415926/EffectData/resolve/main/Videos/${EFFECT_NAME}.zip?download=true"
```

## 📖 Citation

If you find our work / dataset useful, please consider citing:

```bibtex
@article{yang2026effectmaker,
  title={EffectMaker: Unifying Reasoning and Generation for Customized Visual Effect Creation}, 
  author={Yang, Shiyuan and Li, Ruihuang and Tao, Jiale and Shao, Shuai and Lu, Qinglin and Liao, Jing},
  year={2026},
  journal={arXiv preprint arXiv:2603.06014},
  url={https://arxiv.org/abs/2603.06014}, 
}
```