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.
ComfyUI-LTXVideo
LTX-specific ComfyUI nodes and workflow support:
Required Models
LTX-2.5
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
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.
