LTX Video Workflow Step-by-Step Guide

LTX Video Workflow

Introduction to LTX Video Model

LTX Video is a revolutionary DiT architecture video generation model with only 2B parameters, featuring:

Real-time Generation: Capable of generating videos faster than real-time playback
High-Quality Output: Smooth video output at 768x512 resolution and 24FPS
Multiple Generation Modes: Supports text-to-video, image-to-video, and video-to-video conversion

Setup Requirements

System Requirements

Python 3.10.5 or higher
CUDA 12.2 or higher
PyTorch >= 2.1.2

ComfyUI Environment

Update ComfyUI First, ensure your ComfyUI is updated to the latest version. If you’re unsure how to update ComfyUI, please refer to How to Update ComfyUI
Install ComfyUI-LTXVideo Custom Node There are two installation methods:

Method 1: Via ComfyUI Manager (Recommended)

Open ComfyUI Manager
Search for “LTXVideo”
Click Install

Method 2: Manual Installation

Navigate to ComfyUI’s custom_nodes directory
Clone the repository:

git clone https://github.com/Lightricks/ComfyUI-LTXVideo

Install dependencies:

pip install -r requirements.txt

If you’re not familiar with plugin installation, please refer to ComfyUI Plugin Installation Guide

Required Models Download

You need to download the following model files:

Model Name	File Name	Installation Path	Download Link
LTX Video Model	`ltx-video-2b-v0.9.safetensors`	`models/checkpoints`	Hugging Face
PixArt Text Encoder	`model-00001-of-00002.safetensors`	`models/text_encoders/PixArt-XL-2-1024-MS/text_encoder`	Hugging Face
T5 Text Encoder	`t5xxl_fp16.safetensors`	`models/text_encoders`	Hugging Face

Note:

The PixArt text encoder requires downloading the complete text_encoder folder contents

The T5 text encoder file is large (approximately 9.79GB), using a download manager is recommended

Workflow Files

Text-to-Video Workflow

Image-to-Video Workflow

Video-to-Video Workflow

LTX Video Usage Limitations

Resolution and Frame Rate

Resolution must be a multiple of 32
Frame rate must be a multiple of 8 + 1 (e.g., 65 frames, 257 frames, etc.)
Recommended resolution should not exceed 720x1280
Recommended frame count should not exceed 257 frames

Prompt Guidelines

Must be in English
The more detailed the prompt, the better
It is recommended to include complete descriptions of scenes, actions, and details

Workflow Usage Tutorial

Basic Node Descriptions

All workflows include the following basic nodes:

Model Loading Node

LTXVLoader: Load the main LTX Video model
- Select the ltx-video-2b-v0.9.safetensors file
LTXVCLIPModelLoader: Load the text encoder
- Select the PixArt-XL-2-1024-MS/text_encoder/model-00001-of-00002.safetensors file
LTXVModelConfigurator: Configure model parameters
- Set basic parameters such as resolution, frame count, and FPS
- Optionally enable conditioning input

Prompt Processing Node

CLIPTextEncode (Positive): Positive prompt encoding
- Use the PixArt encoder to process positive prompts
CLIPTextEncode (Negative): Negative prompt encoding
- Use the PixArt encoder to process negative prompts
CFGGuider: Control the strength of prompt guidance
- Recommended value range: 2-7
- The larger the value, the closer the generated content will be to the prompt description

Sampling Control Node

KSamplerSelect: Select the sampler
- It is recommended to use the Euler sampler
BasicScheduler: Set the number of sampling steps and scheduler
- Step range: 10-25
- Scheduler type: normal
RandomNoise: Generate random noise
- A fixed seed can be set for reproducible results
SamplerCustomAdvanced: Execute the sampling process
- Integrate all sampling-related parameters for final generation

Output Node

VAEDecode: Decode the generated frames
- Use the built-in VAE decoder of LTX Video
VHS_VideoCombine: Combine the final video
- Set output video frame rate, format, and encoding parameters
- Supports previewing the generated video

LTX Video Generation Mode Tutorial

Text-to-Video

Set Basic Parameters In LTXVModelConfigurator:

Resolution: 768x512
Frame Count: 65 (approximately 2.5 seconds)
FPS: 25

Write Prompts

Positive prompts should be as detailed as possible, describing scenes, actions, and details
Negative prompts are recommended to include: “worst quality, inconsistent motion, blurry, jittery, distorted, watermarks”

Adjust Sampling Parameters

Steps: Recommended 20 steps
CFG: Recommended 4-7
Sampler: Euler
Scheduler: Normal

Image-to-Video

In addition to basic settings, you also need:

Prepare Reference Images

Use the LoadImage node to load reference images
Images should ideally match the target resolution ratio

Adjust Conversion Parameters

Lower the CFG value (recommended 3-5) to maintain consistency with the reference image
Sampling steps can be appropriately reduced (15-20)

Video-to-Video

Load Source Video Use the VHS_LoadVideo node:

Set an appropriate frame rate
Choose whether to adjust the resolution

Parameter Tuning

Use a lower CFG (2-4)
Reduce sampling steps (10-15)
Adjust the sigma_shift parameter as needed

LTX Video Optimization Guide

Parameter Optimization

Prompt Optimization
- Use detailed and specific descriptions
- Include descriptions of actions and scene transitions
- Add vocabulary related to cinematography
Performance Optimization
- Reduce resolution appropriately to increase speed
- Decrease frame count for testing
- Use fewer sampling steps
Quality Optimization
- For shaky images: lower the CFG value
- For insufficient details: increase sampling steps
- For unnatural transitions: optimize prompt descriptions

LTX Video Advanced Application Tips

Long Video Production

Generate multiple segments separately
Maintain stylistic consistency through prompts
Use video editing tools for post-production stitching

Style Control

Include specific artistic style descriptions in prompts
Use reference images to guide style
Adjust style strength through CFG values

Action Control

Describe action processes in detail in prompts
Use keyframes as references
Adjust frame rates appropriately for desired effects

LTX Video Examples and Templates

Scene Examples

Simple Scene Transition

Positive Prompt: “A serene lake at sunrise, gentle ripples on the water surface, morning mist slowly rising, birds flying across the golden sky” Sampling Steps: 20 CFG: 4

Complex Action Sequence Positive Prompt: “A professional dancer performing a graceful contemporary dance sequence, flowing movements, dynamic spins and leaps, soft lighting, studio setting” Sampling Steps: 25 CFG: 5

Remember to save your preferred parameter combinations for future use. Through continuous experimentation and adjustment, you will gradually master the usage of LTX Video.

LTX Video Prompt Template

The turquoise waves crash against the dark, jagged rocks of the shore, sending white foam spraying into the air. The scene is dominated by the stark contrast between the bright blue water and the dark, almost black rocks. The water is a clear, turquoise color, and the waves are capped with white foam. The rocks are dark and jagged, and they are covered in patches of green moss. The shore is lined with lush green vegetation, including trees and bushes. In the background, there are rolling hills covered in dense forest. The sky is cloudy, and the light is dim.