Skip to content

LLaVA-Video

Video instruction tuning with a synthetic data set

Visit official site

What LLaVA-Video is

LLaVA-Video is described as a large multimodal model for video instruction tuning built around the synthetic LLaVA-Video-178K data set.

What you can do with it

  • Tune a model on video instructions
  • Work with descriptions and QA tasks
  • Use SlowFast video representation

Application scenarios

  • A researcher trains a video QA model
  • A developer adds video search to an app
  • A firm filters unsuitable video content

Pros & cons

✓ What we like

  • The data set, generation flow and checkpoints are open
  • Visual instruction tuning data can be combined

! What to watch out for

  • Setup follows the provided Training Code and hardware
  • Users download the data set and checkpoints first

FAQ

What is the data set called?

The material names LLaVA-Video-178K.

Which tasks are included?

The material lists video description, open-ended QA and multiple-choice QA.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review