Using ComfyUI with LTX

Use LTX in ComfyUI with visual node-based workflows. This is the recommended way to work with LTX for most users.

Why ComfyUI?

ComfyUI provides a balance of power and ease of use for LTX:

• Visual workflows - See your entire generation pipeline
• Pre-built examples - Start with working templates
• Real-time experimentation - Iterate quickly on parameters
• Advanced features - Access all LoRAs, control models, and tools
• No coding required - Drag, drop, and connect nodes

Installation

Prerequisites

Before you begin setting up an LTX workflow in ComfyUI, make sure you have:

• ComfyUI installed (Download here)
• CUDA-compatible GPU with 32GB+ VRAM
• 100GB+ free disk space for models and cache
• Python 3.12+

Recommended: ComfyUI Manager

This is the easiest method and handles dependencies automatically.

1. Open ComfyUI
2. Click the Templates button
3. Search for “LTX-2.5”
4. Select a template (we recommend starting with Text to Video or Image to Video)
5. Press “Download all” to download the required models
6. Wait for installation to complete
7. Restart ComfyUI

The nodes will appear in your node menu under the “LTXVideo” category.

Alternative: Manual Installation

For more control:

# Navigate to ComfyUI custom nodes directory
cd ComfyUI/custom_nodes
# Clone the repository
git clone https://github.com/Lightricks/ComfyUI-LTXVideo.git
# Install requirements
cd ComfyUI-LTXVideo
pip install -r requirements.txt

For portable/embedded ComfyUI installations:

.\python_embeded\python.exe -m pip install -r .\ComfyUI\custom_nodes\ComfyUI-LTXVideo\requirements.txt

After installation, restart ComfyUI to load the new nodes.

Verifying Installation

After restarting ComfyUI:

1. Right-click in the canvas
2. Navigate to Add Node → LTXVideo
3. You should see categories like:
   • LTXVideo/loaders
   • LTXVideo/samplers
   • LTXVideo/conditioning
   • LTXVideo/utils

If you don’t see these, check the ComfyUI console for error messages.

Models

Automatic Download (Recommended)

The LTX nodes will automatically download required models on first use. Simply load a workflow and click “Queue Prompt” - the nodes will handle downloading.

Manual Download

If you prefer to download models manually or need offline installation:

Standard model files (BF16)

Download the LTX-2.5 files from the LTX-2.5 HuggingFace repository and place each in the matching ComfyUI folder:

The spatial upscaler is only used by the two-stage templates (Text-to-Video, Image-to-Video); the single-stage First-Frame/Last-Frame template doesn’t need it.

Lower-VRAM model files

For a lower-VRAM local setup with the distilled LTX-2.5 video workflows, use these three files together in place of the BF16 transformer, Gemma text encoder, and video VAE above:

Text-to-Audio has no video branch, so it uses only the INT8 ConvRot transformer and text encoder together with the BF16 audio VAE — not the Conv video VAE.

New in the LTX-2.5 templates

Separate audio and video VAEs. LTX-2.5 uses separate VAEs for video and audio: the video VAE decodes the generated video frames, and the audio VAE (ltx-2.5-audio-vae-bf16.safetensors) decodes the generated soundtrack. Use either the higher-quality DiffVAE (ltx-2.5-video-vae-bf16.safetensors) or the faster, lighter Conv VAE (ltx-2.5-video-vae-conv-bf16.safetensors) according to the configuration above. The official Text-to-Video and Image-to-Video templates use both video and audio VAEs.

Text encoder and prompt enhancer. The templates use two Gemma 4 files with different roles:

• Text encoder — converts the final prompt into conditioning for the LTX-2.5 model. Use gemma4-12b-with-proj-ltx-2.5-bf16.safetensors for the reference BF16 configuration or gemma4-12b-with-proj-ltx-2.5-comfy-int8-convrot.safetensors for the lower-VRAM ComfyUI configuration. One compatible text encoder file is required.
• Prompt enhancer — gemma4_e2b_it_bf16.safetensors can expand a short prompt into a more detailed one before it’s encoded. The supplied templates enable prompt enhancement by default; turn off Prompt Enhance to use your original prompt without rewriting.

Loading and Using Workflows

Loading a Workflow

1. In ComfyUI, click Workflows
2. Navigate to custom_nodes/ComfyUI-LTXVideo/example_workflows/
3. Select a .json workflow file
4. Click Open

Modifying Parameters

Click any node to see its parameters in the sidebar:

1. Text prompts: Click text encode nodes
2. Generation settings: Click sampler nodes
3. Output settings: Click save nodes

Running a Workflow

1. Configure your prompt and settings
2. Click Queue Prompt (or press Ctrl+Enter)
3. Watch the progress bar
4. Find output in your ComfyUI/output/ folder

Installing Missing Nodes

When you load a workflow with missing nodes:

1. ComfyUI will show a warning dialog
2. Click Install Missing Custom Nodes
3. ComfyUI Manager will install automatically
4. Restart ComfyUI after installation

Or install manually:

cd ComfyUI/custom_nodes
# Replace the URL below with the missing node's GitHub repository
git clone https://github.com/Lightricks/ComfyUI-LTXVideo.git
cd ComfyUI-LTXVideo
pip install -r requirements.txt

Templates and Example Workflows

Built-in templates

ComfyUI includes three ready-made LTX-2.5 templates:

• Text-to-Video (video_ltx2_5_t2v) — two-stage; generate from a text prompt. Walkthrough: Text-to-Video.
• Image-to-Video (video_ltx2_5_i2v) — two-stage; generate from a starting image. Walkthrough: Image-to-Video.
• First-Frame / Last-Frame (video_ltx2_5_flf2v) — single-stage; generates the motion between a start and end frame.

Press Download all when a template loads to fetch the model files it needs.

Advanced example workflows

The ComfyUI-LTXVideo repo hosts advanced workflows — IC-LoRA control, in/outpainting, and more — under example_workflows/2.5/.

Workflow Tips

Starting a new project:

1. Load a basic workflow first
2. Test with low resolution (480×720)
3. Use fewer frames (41-81) for speed
4. Increase quality once satisfied

Memory management:

• Close unused workflows
• Clear cache between large generations
• For VRAM savings, use the quantized lower-VRAM model files (INT8 ConvRot) instead of the BF16 stack; an even smaller NVFP4 distilled transformer is available for FP4-capable (NVIDIA Blackwell) GPUs
• Process in batches if generating multiple videos

Troubleshooting

Nodes Not Appearing

Issue: LTXVideo nodes don’t show up after installation

Solutions:

• Verify installation in ComfyUI/custom_nodes/ComfyUI-LTXVideo/
• Check ComfyUI console for errors during startup
• Reinstall requirements: pip install -r requirements.txt
• Restart ComfyUI completely (close and reopen, not just refresh)
• Try manual installation method if ComfyUI Manager failed
• Check Python version (should be 3.12+)

Workflow Errors

Issue: Workflow loads but shows red nodes or errors

Solutions:

• Missing nodes: Click “Install Missing Custom Nodes” in the error dialog
• Update ComfyUI: Make sure you’re on the latest version
• Update LTXVideo nodes:

  cd ComfyUI/custom_nodes/ComfyUI-LTXVideo
  git pull
  pip install -r requirements.txt --upgrade
• Check model paths: Verify models are in correct locations
• Incompatible workflow: Some older workflows may need updates for newer node versions

Missing Dependencies

Issue: Workflow requires additional custom nodes

Solutions:

• ComfyUI will show a list of missing nodes when loading
• Click “Install Missing Custom Nodes” button
• Common dependencies:
  • ComfyUI-VideoHelperSuite (video processing)
  • ComfyUI-Manager (node management)
  • ComfyUI-Impact-Pack (advanced features)
• Restart ComfyUI after installing dependencies

Model Loading Errors

Issue: Models fail to load or show errors in console

Solutions:

• Verify model files aren’t corrupted (check file sizes)
• Ensure models are in correct directories:
  • Diffusion model: ComfyUI/models/diffusion_models/
  • Text encoders (Gemma 4): ComfyUI/models/text_encoders/
  • VAE (video + audio): ComfyUI/models/vae/
  • Spatial upscaler: ComfyUI/models/latent_upscale_models/
  • LoRAs / IC-LoRAs: ComfyUI/models/loras/
• Try re-downloading the model
• Check that you have the correct model version for your nodes
• Look for specific error messages in the ComfyUI console

Advanced Features

Custom Workflow Creation

Build your own workflows by combining nodes:

1. Start with a template
2. Add/remove nodes as needed
3. Connect nodes with click-and-drag
4. Save your workflow for reuse

LTX offers a number of custom nodes you can use. See a list here.

Common node patterns:

Basic generation:

Model Loader → Text Encode → Sampler → VAE Decode → Save

Image-to-video:

Load Image → Image Conditioning → Sampler → VAE Decode → Save

With upscaling:

Sampler → Upscaler → Sampler → VAE Decode -> Save

Getting Help