Skip to content

Dataset · WenhaoWang

VidProM: 1.67M prompts and 6.69M videos

WenhaoWang/VidProM

Million-scale dataset of text prompts and generated videos for text-to-video research.

What's inside

Contains 1.67M unique text prompts for video generation and 6.69M videos created by four modern diffusion models. Language is English; metadata also includes tabular data and videos, with a volume of 1k to 10k records. This is the first dataset of this scale for text-to-video prompts, presented at NeurIPS 2024.

What you can do with it

  • Text-to-video prompt engineering
  • Training and evaluating fake video detectors
  • Analyzing video generation efficiency
Tasks
Text → video, Text → image
Languages
English
Size
1M–10M rows
Downloads this month
22.8K
Likes
84

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("WenhaoWang/VidProM")