How To Use FramePack (Gradio) To Create Smooth, 120 Seconds AI Videos Locally

Have you ever wanted to create long, smooth AI-generated videos on your own computer, without needing a powerful server or a high-end GPU? Meet FramePack, a revolutionary new video generation framework that makes this possible.

In this post, I’ll walk you through how FramePack works, what makes it special, and how to run it locally using Gradio.

What Is FramePack?

Frame Pack is an AI video generation system developed by researchers Alvin Zhang and Manish Agrawal. If you’re into open-source AI, you might recognize Alvin from the popular ControlNet and IC-Light projects.

Unlike traditional video generation models that consume massive GPU resources as video length increases, Frame Pack is built for efficiency. It can run smoothly on consumer hardware, even with just 6GB of VRAM—no data center or monster GPU required!

Why FramePack Is a Game Changer

Most video models get slower and heavier with longer clips. That’s because they process every frame the same way, overloading your system. FramePack flips the script using a technology called:

Constant-Length Context Compression

This technique keeps your PC’s workload constant, no matter how long your video is. It’s like turning a marathon into a sprint for your GPU.

Bidirectional Sampling

This solves a common AI video issue known as drift—where animations get weird over time—by generating frames both forwards and backwards to ensure visual consistency.

Patchifying Kernel

This smartly allocates more computing power to areas of the video that need it, producing higher-quality results without wasting resources.