In the ever-evolving landscape of AI video generation, a new open-source contender has emerged that promises remarkable results with unparalleled speed - and it’s completely free. Meet LTX Video (LTXV), developed by Litrix, now updated to version 0.9.7 and powered by a robust 13B parameter model. With multiscale rendering, high-resolution support, and offline capability, LTXV is shaping up to be one of the most accessible and powerful AI video tools currently available.

In this blog post, we’ll explore what makes LTX Video exceptional, how you can use its online and offline versions, and walk you through the complete local installation process using ComfyUI.

🚀 Why LTX Video Stands Out

• Free & Open Source
  LTXV is entirely open source and free to use, whether online or offline.

• Blazing Fast
  Claimed to be up to 30× faster than competitors - and real-world usage confirms impressive performance.

• High Quality
  Generates full HD videos with detailed rendering through its progressive multiscale approach.

• Consumer-Grade GPU Compatibility
  Works efficiently on GPUs with as little as 8 GB VRAM using the quantized version.

• Key Features:

  • Progressive video generation via multiscale layers.
  • Free spatial and temporal upscaling models.
  • Keyframe-based video generation using start/middle/end frames.
  • Integration with ComfyUI for powerful offline workflows.
  • Frame interpolation, camera movement control, and more.

🌐 Online Use via LTX Studio

If you prefer a browser-based experience without using your own GPU, head over to LTX Studio. Features include:

• Storyboard creation
• Image generator
• Motion Generator with 13B model
• Start/end image frame support
• Intuitive camera movement presets
• Automatic sound effect generation
• Paid upscaling options for higher resolution output

💻 Running LTX Video Offline (Using ComfyUI)

Running LTXV locally offers complete freedom and unlimited generation without paid restrictions. Follow these steps to get set up:

Pre-requisites

• ComfyUI
• Git / GitHub
• Python 3.11.0

🛠️ Installing ComfyUI

Before using LTX Video offline, you need to install ComfyUI, a powerful and flexible graphical interface for running AI image and video workflows.

Follow these steps to install ComfyUI:

1. Clone the repository:

   git clone https://github.com/comfyanonymous/ComfyUI.git
   cd ComfyUI

2. (Windows) Install dependencies and run:

   python -m venv venv
   venv\Scripts\activate
   pip install -r requirements.txt
   python main.py

3. (Linux/macOS) Install dependencies and run:

   python3 -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt
   python3 main.py

🧰 Installing ComfyUI Manager (Required Step)

To enable custom node management for installing LTXV components, you must install ComfyUI Manager:

1. Open your terminal or command prompt.

2. Navigate to the custom_nodes directory inside your ComfyUI installation:

   cd path/to/ComfyUI/custom_nodes

3. Clone the ComfyUI Manager repository:

   git clone https://github.com/ltdrdata/ComfyUI-Manager comfyui-manager

4. Restart ComfyUI to load the new manager.

After this, you can access Manager > Custom Nodes Manager within the ComfyUI interface to install required extensions like LTX.

Once ComfyUI is running, you can proceed with installing LTXV and the necessary workflows.

2. Download LTXV Model (Quantized FP8 by KI)

Avoid the bulky Q8 model which requires Microsoft Build Tools. Instead:

• Download the FP8 quantized model by KI (~15 GB).
• Save it in:
  ComfyUI/models/checkpoints

3. 🧬 Downloading T5XXL_FP8 Clip Model

To ensure proper prompt processing in your ComfyUI LTXV workflows, download the required clip model:

1. Go to the following Hugging Face repository:
   https://huggingface.co/lllyasviel/flux_text_encoders/tree/main

2. Download the file:
   t5xxl_fp8_e4m3fn.safetensors

3. Save it into your ComfyUI directory under:
   ComfyUI/models/clip/

This model is required for the text encoding component of your workflows and ensures compatibility with the FP8 quantized LTXV model.

4. Download the Workflow File

Download an LTXV image-to-video workflow JSON file and save it anywhere.

5. Update ComfyUI

Launch ComfyUI, go to Manager > Update ComfyUI to get the latest node compatibility.

6. Drag Workflow onto Canvas

• Drag the downloaded workflow JSON into the ComfyUI canvas.
• If missing nodes appear, go to Manager > Custom Nodes Manager and install the LTX node package.
• Restart ComfyUI.

7. Load Model and Prompt

In the workflow:

• Load T5XXL_FP8 clip model (download from HuggingFace if not available).
• Set positive/negative prompts as needed.
• Select the FP8 quantized LTXV model.
• Set input image as the first frame.
• Configure resolution (default: 768×512) and video length.

8. Generate & Save

• Run the workflow.
• Right-click the preview and choose “Save Preview” to store your video.

🔼 Upscaling for Full HD Output

LTXV supports post-generation upscaling:

• Spatial Upscaler – Increases pixel resolution.
• Temporal Upscaler – Improves temporal consistency between frames.

Steps:

• Download LTXV Spatial Upscaler from the repo and save in:
  ComfyUI/models/upscale_models
• Enable upscaler in workflow (set to true) before running.

🎞️ Keyframe Workflow (Optional)

You can add start, middle, and end images to guide video flow:

• Download the keyframe workflow JSON file.
• Add up to three images.
• Write a transition prompt (e.g., “The shoe hovers, blue bolts swirl, then flames ignite”).
• Run and upscale as desired.

🧠 Final Thoughts

LTX Video 13B (v0.9.7) is a game-changing AI video generator. Combining open-source accessibility, exceptional speed, and quality outputs, it’s a compelling choice for creators and developers alike. The ability to run it offline - even on consumer-grade hardware - puts powerful generative tools into more hands than ever before.

Whether you're experimenting with artistic animations, storytelling sequences, or cinematic effects, LTXV gives you an exciting new canvas to explore.

🔗 Useful Resources

• LTX Studio
• ComfyUI GitHub
• FP8 Model by KI
• Spatial & Temporal Upscalers
• T5XXL FP8 Clip Model