Install LTX Video: The Fastest Local AI Video Generator for ComfyUI on Windows

LTX Video is an innovative tool by Lightricks that brings the power of AI video generation directly to your local machine. Designed to integrate seamlessly with ComfyUI, LTX Video offers unparalleled speed and efficiency, making it an excellent choice for creative professionals and hobbyists alike. In this guide, we’ll walk you through the step-by-step process of installing LTX Video on Windows, so you can unlock the fastest local AI video generation experience today!

Requirements

• ComfyUI with ComfyUI Manager
• Nvidia GPU with 16GB of VRAM, 24GB of VRAM or more
• git and git-lfs

Installation

• Click on Manager on the upper right corner of ComfyUI

  

• Click on Custom Nodes Manager
  

• Search for LTX and click on Install to install ComfyUI-LTXVideo node

  

• Click on Restart to restart ComfyUI. You might need to restart it several times for the installaiton to be complete.

  

Download the Models

• Download ltx-video-2b-v0.9.safetensors from Hugging Face and place it under models\checkpoints
• Change directory to models\text_encoders and type in this command:

  git clone https://huggingface.co/PixArt-alpha/PixArt-XL-2-1024-MS
  

• Restart or refresh ComfyUI after the download is finished.

Samples

• Text to Video (drag the fullsize picture to the ComfyUI canvas)
  
  
• Image to Video
  
  
• Video to Video
  I don’t have an example. Here is the link to the workflow. Note that the default workflow might have the wrong text encoder selected, you need to re-select the text encoder.

Tips

• The resolution 768×512 works the best.
• If the video comes out static, you might need to process your image to add compression. I will have a workflow for that.
• Detailed prompts works better. You can use ChatGPT to come up with the prompt. You can see my post on how to use ChatGPT for prompts. It’s for Flux model, but it still applies in this case.
• The success rate of getting good results is still low, but the generation speed is fast. You can queue it to generate several videos and pick from the batch.