Configure LTX 2.3 in ComfyUI

Step-by-step

1. Select the downloaded model in the GGUF model loader node
2. Press the R key on your keyboard to refresh node definitions if you added models while ComfyUI was running
3. Connect the Dual Clip text encoder to transform prompts into latent data
4. Configure the Video VAE and Audio VAE nodes to decode the latent space into pixels and sound
5. Set the video resolution to a value divisible by 32
6. Calculate the frame count by multiplying the desired seconds by the FPS and adding 1 (e.g., 121 for 5 seconds at 24 FPS)
7. Enter the generation subgraph to monitor the sampling stages
8. Define the output file path and naming convention in the Save node

Options

• Use Distilled LoRA to speed up workflow and improve quality
• Add additional LoRA nodes for custom style adjustments
• Adjust resolution up to Full HD depending on GPU capacity

Watch out for

• Resolution must be divisible by 32 to avoid blurry edges or artifacts
• If a model loader says 'undefined', the model file is in the wrong folder
• The frame count must include an extra frame (result + 1) to prevent the last frame from looking weird due to temporal rendering

Tips

• Start with a low resolution if you have an older graphics card
• Use the GGUF model loader for the primary video model as it handles the heavy lifting
• Organize sampling stages into subgraphs to make the workflow visually easier to manage