Skip to content

CelebV-Text

A face text-video dataset for generation research

Visit official site

What CelebV-Text is

The dataset is described as a large, diverse collection of face video clips with paired texts, built to support face text-to-video generation research.

What you can do with it

  • Run face text-to-video generation research
  • Analyse correlation between text and video
  • Build a benchmark for the generation task

Dataset notes

  • 70,000 in-the-wild face video clips are included
  • Each clip carries 20 texts
  • Labels cover appearance, lighting, action, emotion and light direction

Pros & cons

✓ What we like

  • Full statistical analysis of video, text and their correlation
  • Diverse labels for appearance, lighting, action and emotion

! What to watch out for

  • The material centres on research use of the dataset
  • Evaluation is framed around the built benchmark

FAQ

How large is the dataset?

The material states 70,000 clips with 20 texts each.

What is included beyond clips?

The material lists statistical analysis and a standardised benchmark.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review