Skip to content

Pyramid Flow

Flow matching for autoregressive video generation

Visit official site

What Pyramid Flow is

The framework is described as a flow matching video method built on an autoregressive model and developed by three named institutions.

What you can do with it

  • Generate video from a text description
  • Generate video from a text-conditioned image
  • Try the interactive Hugging Face demo

Getting started

  • Fetch the code from the GitHub page
  • Install dependencies following the README guide
  • Load pretrained models from Hugging Face

Pros & cons

✓ What we like

  • Training is stated at 20.7k A100 GPU hours
  • Text and image conditioning are both offered

! What to watch out for

  • Setup follows the README install steps
  • Output uses the provided scripts and CLI tools

FAQ

Who developed it?

The material names Peking University, Kuaishou Technology and Beijing University of Posts and Telecommunications.

Where is the code?

The material lists GitHub code and Hugging Face pretrained models.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review