LTX-2.5 INT8 T2V, I2V, FLF & Video Extension Workflow

An optimized ComfyUI workflow for LTX-2.5, completely redesigned from the previous LTX-2.3 workflows.

The workflow uses INT8 inference and supports multiple generation and editing modes, including Text-to-Video, Image-to-Video, First-Last-Frame, video extension, custom audio, prompt relay, and voice cloning.

It also provides selectable Distilled / Full model modes and 25 / 50 FPS output.

Workflow

Workflow	File	Description
LTX-2.5 INT8	`LTX2.5_int8_I2V-javanoYYMM.x.json`	Unified T2V, I2V, FLF, Extend, Custom Audio, Prompt Relay, and Voice Cloning workflow.

Features

INT8

Uses an INT8 version of LTX-2.5 for efficient local inference.

Benefits:

Reduced VRAM usage
Faster inference
Improved memory efficiency
Better suitability for consumer GPUs

T2V / I2V

Supports both:

Text-to-Video (T2V)
Image-to-Video (I2V)

Generate video directly from a text prompt or use an input image as the visual starting point.

First-Last-Frame (FLF)

Generate controlled transitions between a first and last frame.

Features:

First-frame conditioning
Last-frame conditioning
Controlled transitions
Character and scene continuity
Camera movement
Transformation sequences

Video Extend

Extend video sequences from either the beginning or the end.

The workflow supports both:

LTX-2.5 generated videos
Existing videos

This allows longer sequences to be created iteratively while maintaining temporal and audio continuity.

Custom Audio

Use externally supplied audio as part of the generation process.

Supported use cases include:

Dialogue
Voice-over
Music
Sound effects
Existing video audio

This allows video generation to be synchronized to predetermined audio.

Prompt Relay

Prompt Relay allows prompt information to be passed between generation stages.

Useful for:

Video extension
Multi-stage generation
Maintaining scene descriptions
Iterative video generation
Long-video workflows

Voice Cloning

Supports voice cloning using a reference voice.

Useful for:

Character dialogue
Talking videos
Narration
Voice-over
Consistent character voices

Distilled / Full

The workflow supports switching between Distilled and Full model configurations.

Distilled

Faster generation
Fewer sampling steps
Rapid iteration

Full

Higher generation quality
Greater fidelity
Final rendering

25 / 50 FPS

The workflow supports selectable output frame rates:

25 FPS

Lower computational requirements
Faster generation
Standard video workflows

50 FPS

Smoother motion
High-frame-rate output
Better suited to motion-intensive scenes

Workflow Details

`LTX2.5_int8_I2V-javanoYYMM.x.json`

A unified LTX-2.5 INT8 workflow supporting multiple generation modes.

Generation

Extension

Extend from the beginning
Extend from the end
Extend LTX-2.5 generated videos
Extend existing videos

Audio

Native audio generation
Custom Audio
Voice Cloning
Audio/video synchronization

Control

Prompt Relay
First/Last Frame conditioning
Distilled / Full switching
25 / 50 FPS switching

Required Software

ComfyUI

Use a recent version of ComfyUI with LTX-2.5 support.

https://github.com/comfyanonymous/ComfyUI

ComfyUI-LTXVideo

LTX-specific ComfyUI nodes and workflow support:

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

Required Models

LTX-2.5

Lightricks/LTX-2.5

https://huggingface.co/Lightricks/LTX-2.5

The workflow uses the required LTX-2.5 INT8 model components.

Additional text encoder, VAE, audio, or voice-related models should be downloaded according to the workflow requirements.

Installation

Update ComfyUI.
Install or update ComfyUI-LTXVideo if required.
Download the required LTX-2.5 INT8 model files.
Install the required text encoder, VAE, and audio components.
Install any missing custom nodes reported when loading the workflow.
Load:

LTX2.5_int8_I2V-javanoYYMM.x.json

Select the desired generation mode and configuration.
Choose Distilled or Full.
Choose 25 FPS or 50 FPS.
Generate.

Recommended Configuration

Fast Generation:
INT8 + Distilled + 25 FPS

High Quality:
INT8 + Full + 50 FPS

Video Extension:
INT8 + Distilled/Full + Prompt Relay + Extend

Custom Audio:
INT8 + Custom Audio

Voice Cloning:
INT8 + Voice Cloning + Custom Audio

The optimal configuration depends on GPU architecture, VRAM, resolution, video length, sampling settings, and model mode.

Notes

T2V = Text-to-Video.
I2V = Image-to-Video.
FLF = First-Last-Frame.
Extend supports both LTX-2.5 generated videos and existing videos.
Prompt Relay is designed for multi-stage and extended video generation.
Custom Audio allows externally supplied audio to drive the generation process.
Voice Cloning requires the appropriate reference voice and supporting models.
Distilled mode prioritizes speed.
Full mode prioritizes generation quality.
25 FPS requires less computation than 50 FPS.
50 FPS provides smoother motion but requires additional computation.
Actual performance depends on GPU architecture, VRAM, resolution, frame count, and model configuration.

Additional Resources

LTX-2.5
https://huggingface.co/Lightricks/LTX-2.5

LTX-2
https://github.com/Lightricks/LTX-2

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

LTX Documentation
https://docs.ltx.video/

License

Please follow the respective licenses of LTX-2.5, ComfyUI, ComfyUI-LTXVideo, voice-cloning models, text encoders, audio models, and any other third-party models or custom nodes used by this workflow.

Downloads last month: -; Downloads are not tracked for this model. How to track