Training LTX-2

The LTX-2 Trainer enables researchers and advanced users to fine-tune the LTX-2 model on custom datasets. You can train LoRAs, perform full fine-tuning, or create video-to-video transformations (IC-LoRA).

Who Should Use the Trainer?

The trainer is designed primarily for:

• Researchers exploring video generation model training techniques
• Advanced users building specialized video generation models

Requirements

Before starting, ensure you have:

• LTX-2 Model Checkpoint - Local .safetensors file
• Gemma Text Encoder - Local Gemma model directory (download from HuggingFace)
• Linux with CUDA - CUDA 13+ recommended
• Nvidia H100 GPU with 80GB+ VRAM - Lower VRAM may work with gradient checkpointing and reduced resolutions

Additional Documentation

For comprehensive training documentation, including setup instructions, dataset preparation guides, configuration options, and troubleshooting, visit the LTX-2 Trainer GitHub repository.

The GitHub documentation includes:

• Quick Start Guide - Get training in minutes
• Dataset Preparation - Video preprocessing and captioning
• Training Modes - Detailed explanations of each training approach
• Configuration Reference - All available training parameters
• Troubleshooting - Solutions to common issues
• Utility Scripts - Tools for dataset management