LTX Video

Coding Agents 💻 Python ⚖️ Apache-2.0 🔴 No recent commits
10.9k stars

LTX-Video is a DiT-based video generation model that generates high-fidelity videos with synchronized audio, supporting up to 4K resolution and 50 FPS. It solves the need for a unified model that handles text-to-video, image-to-video, and video extension tasks. It is for developers and creators who want to generate or edit videos programmatically or via ComfyUI.

LTX Video demo
🎞️ Demo from the project README

✨ Key features

  • Generates synchronized audio and video in one pass
  • Supports up to 4K resolution and 50 FPS
  • Image-to-video, multi-keyframe, and video extension
  • Distilled models for faster inference and lower VRAM
  • Control models for depth, pose, and canny conditioning
  • ComfyUI and Diffusers integration

🎯 Use cases

  • Generate videos from text prompts
  • Animate still images with motion and audio
  • Extend existing videos forward or backward
  • Create controlled videos using depth, pose, or canny inputs
  • Rapid prototyping with distilled models for real-time generation

📦 Installation

🧰 Requirements: Requires Python and PyTorch; GPU recommended for full performance, but MPS on macOS is supported. Model weights are downloaded from HuggingFace.

Clone the repository
git clone https://github.com/Lightricks/LTX-Video.git
cd LTX-Video
Install dependencies
pip install -r requirements.txt

For ComfyUI, install the custom node from the ComfyUI Manager or manually clone the repository into the custom_nodes folder.

🚀 Usage

# Example inference using the official script
python inference.py \
  --config configs/ltxv-13b-0.9.8-dev.yaml \
  --ckpt path/to/ltxv-13b-0.9.8-dev.safetensors \
  --prompt "A cat playing piano" \
  --output output.mp4

For image-to-video, add --image_path input.png.

⚠️ Good to know

The repository is in active development; LTX-2 is the new primary model with improved features, and LTX-Video may have limited support going forward.

❓ FAQ

What is the difference between the base and distilled models?

Distilled models are optimized for faster inference with fewer diffusion steps and do not require classifier-free guidance or spatio-temporal guidance, but may have slight quality reduction.

Can I use LTX-Video with ComfyUI?

Yes, LTX-Video is integrated into ComfyUI, and example workflows are provided in the repository.

What are the system requirements?

A GPU is recommended for reasonable performance, but MPS on macOS is supported with PyTorch 2.3.0. The 13B models require more VRAM, while the 2B models are lighter.

Is there a newer version of LTX-Video?

Yes, LTX-2 is the next generation with synchronized audio+video and is now the primary home for LTX development.

📊 Repository

Stars★ 10,933
Forks🍴 1,125
Open issues🐛 100
Last commit🕒 Jan 5, 2026
Created📅 Nov 2024
Language💻 Python
License⚖️ Apache-2.0

🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.