Getting Started with LTX 2.3: Complete Tutorial

LTX 2.3 is Lightricks' latest text-to-video model, offering unprecedented quality and control. This tutorial will guide you through the complete workflow from installation to rendering your first video.

Prerequisites

Before starting, ensure you have:

• NVIDIA GPU with at least 12GB VRAM (16GB+ recommended)
• ComfyUI installed and configured
• Basic understanding of AI video generation concepts

Step 1: Download Required Models

Visit our Models page and download:

# Required files (all VRAM tiers)
ltx-video-2b-v0.9.safetensors  # Main checkpoint
vae_diffusion_pytorch_model.safetensors  # VAE decoder
spatial_upscaler_v0.1.safetensors  # Spatial upscaler

For VRAM-optimized versions, check the VRAM guide.

Step 2: Install ComfyUI Nodes

LTX 2.3 requires specific custom nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Lightricks/ComfyUI-LTXVideo
cd ComfyUI-LTXVideo
pip install -r requirements.txt

Restart ComfyUI after installation.

Step 3: Load the Workflow

Download our starter workflow and import it into ComfyUI:

1. Open ComfyUI in your browser
2. Click "Load" button
3. Select the downloaded JSON file
4. Verify all nodes are green (no missing dependencies)

Step 4: Configure Your Prompt

The prompt is crucial for quality results. Follow these guidelines:

Good prompts:

• Specific and descriptive
• Include camera movement (pan, zoom, dolly)
• Mention lighting and atmosphere
• Specify subject actions

Example:

A majestic eagle soaring through misty mountain peaks at golden hour,
camera slowly panning right, cinematic lighting, 4K quality

Step 5: Adjust Generation Settings

Key parameters to tune:

Step 6: Generate Your Video

1. Click "Queue Prompt" in ComfyUI
2. Monitor VRAM usage in Task Manager
3. Wait for generation (2-5 minutes depending on settings)
4. Find output in ComfyUI/output/ folder

Troubleshooting

Out of Memory Error:

• Reduce resolution to 512x512
• Lower frame count to 97
• Use FP8 quantized models
• Enable --lowvram flag in ComfyUI launch

Poor Quality Results:

• Increase steps to 40+
• Adjust CFG scale (try 8-10)
• Refine prompt with more details
• Check model files are not corrupted

Next Steps

Now that you've generated your first video:

• Experiment with advanced workflows
• Try LoRA fine-tuning
• Join the community on Discord for tips

Happy creating!