🚀 New: LTX-2 is Now Available!
We're excited to announce LTX-2 - the next generation of LTX with synchronized audio+video generation!
LTX-2 is the first DiT-based audio-video foundation model that contains all core capabilities of modern video generation in one model. LTX-2 is now the primary home for LTX development and includes significant improvements:
- 🎵 Synchronized Audio+Video Generation - Generate videos with perfectly synchronized audio
- 🎬 Latest Model - LTX-2 with improved quality and capabilities
- 🔌 ComfyUI Integration - Built into ComfyUI core for seamless workflows
- 🎯 Advanced Features:
- Multiple keyframe support
- IC-LoRA control models for precise generation
- Standard LoRA support for style customization
- Latent upsampler for multiscale pipelines
- 🛠️ Training Tools - LoRA training capabilities
- 📚 Comprehensive Documentation - Full documentation at https://docs.ltx.video
- 🔄 Active Development - Ongoing improvements and community support
👉 Check out LTX-2 here
📖 View Documentation
Table of Contents
Introduction
LTX-Video is the first DiT-based video generation model that contains all core capabilities of modern video generation in one model: synchronized audio and video, high fidelity, multiple performance modes, production-ready outputs, API access, and open access. It can generate up to 50 FPS videos at native 4K resolution with synchronized audio in one pass.
The model is trained on a large-scale dataset of diverse videos and can generate high-resolution videos with realistic and diverse content.
The model supports image-to-video, multi-keyframe conditioning, keyframe-based animation, video extension (both forward and backward), video-to-video transformations, and any combination of these features.
Image-to-video examples
Controlled video examples
News
October 23, 2025: LTX-2 Announced
Today we announced our newest foundation model, LTX-2. LTX-2 represents a major leap forward from our previous model, LTXV 0.9.8. Here’s what’s new:
- Audio + Video, Together: Visuals and sound are generated in one coherent process, with motion, dialogue, ambience, and music flowing simultaneously.
- 4K Fidelity: Professional-grade precision with native 4K and up to 50 fps, sharp textures, clean motion, and synchronized audio.
- Longer Generations: LTX-2 supports longer, continuous clips with synchronized audio up to 10 seconds.
- Low Cost & Efficiency: Up to 50% lower compute cost than competing models, powered by a multi-GPU inference stack.
- Creative Control: Multi-keyframe conditioning, 3D camera logic, and LoRA fine-tuning deliver frame-level precision and style consistency.
For more details, please see our blog post. LTX-2 model weights, code, and benchmarks will be released to the community later in 2025.
July, 16th, 2025: New Distilled models v0.9.8 with up to 60 seconds of video:
- Long shot generation in LTXV-13B!
- LTX-Video now supports up to 60 seconds of video.
- Compatible also with the official IC-LoRAs.
- Try now in ComfyUI.
- Release a new distilled models:
- 13B distilled model ltxv-13b-0.9.8-distilled
- 2B distilled model ltxv-2b-0.9.8-distilled
- Both models are distilled from the same base model ltxv-13b-0.9.8-dev and are compatible for use together in the same multiscale pipeline.
- Improved prompt understanding and detail generation
- Includes corresponding FP8 weights and workflows.
- Release a new detailer model LTX-Video-ICLoRA-detailer-13B-0.9.8
July, 8th, 2025: New Control Models Released!
- Released three new control models for LTX-Video on HuggingFace:
May, 14th, 2025: New distilled model 13B v0.9.7:
- Release a new 13B distilled model ltxv-13b-0.9.7-distilled
- Amazing for iterative work - generates HD videos in 10 seconds, with low-res preview after just 3 seconds (on H100)!
- Does not require classifier-free guidance and spatio-temporal guidance.
- Supports sampling with 8 (recommended), or less diffusion steps.
- Also released a LoRA version of the distilled model, ltxv-13b-0.9.7-distilled-lora128
- Requires only 1GB of VRAM
- Can be used with the full 13B model for fast inference
- Release a new quantized distilled model ltxv-13b-0.9.7-distilled-fp8 for real-time generation (on H100) with even less VRAM
May, 5th, 2025: New model 13B v0.9.7:
- Release a new 13B model ltxv-13b-0.9.7-dev
- Release a new quantized model ltxv-13b-0.9.7-dev-fp8 for faster inference with less VRam
- Release a new upscalers
- Breakthrough prompt adherence and physical understanding.
- New Pipeline for multi-scale video rendering for fast and high quality results
April, 15th, 2025: New checkpoints v0.9.6:
- Release a new checkpoint ltxv-2b-0.9.6-dev-04-25 with improved quality
- Release a new distilled model ltxv-2b-0.9.6-distilled-04-25
- 15x faster inference than non-distilled model.
- Does not require classifier-free guidance and spatio-temporal guidance.
- Supports sampling with 8 (recommended), or less diffusion steps.
- Improved prompt adherence, motion quality and fine details.
- New default resolution and FPS: 1216 × 704 pixels at 30 FPS
- Still real time on H100 with the distilled model.
- Other resolutions and FPS are still supported.
- Support stochastic inference (can improve visual quality when using the distilled model)
March, 5th, 2025: New checkpoint v0.9.5
- New license for commercial use (OpenRail-M)
- Release a new checkpoint v0.9.5 with improved quality
- Support keyframes and video extension
- Support higher resolutions
- Improved prompt understanding
- Improved VAE
- New online web app in LTX-Studio
- Automatic prompt enhancement
February, 20th, 2025: More inference options
- Improve STG (Spatiotemporal Guidance) for LTX-Video
- Support MPS on macOS with PyTorch 2.3.0
- Add support for 8-bit model, LTX-VideoQ8
- Add TeaCache for LTX-Video
- Add ComfyUI-LTXTricks
- Add Diffusion-Pipe
December 31st, 2024: Research paper
December 20th, 2024: New checkpoint v0.9.1
- Release a new checkpoint v0.9.1 with improved quality
- Support for STG / PAG
- Support loading checkpoints of LTX-Video in Diffusers format (conversion is done on-the-fly)
- Support offloading unused parts to CPU
- Support the new timestep-conditioned VAE decoder
- Reference contributions from the community in the readme file
- Relax transformers dependency
November 21th, 2024: Initial release v0.9.0
- Initial release of LTX-Video
- Support text-to-video and image-to-video generation
Models
Quick Start Guide
Online inference
The model is accessible right away via the following links:
Run locally
Installation
The codebase was tested with Python 3.10.5, CUDA version 12.2, and supports PyTorch >= 2.1.2.
On macOS, MPS was tested with PyTorch 2.3.0, and should support PyTorch == 2.3 or >= 2.6.
git clone https://github.com/Lightricks/LTX-Video.git
cd LTX-Video
# create env
python -m venv env
source env/bin/activate
python -m pip install -e .\[inference\]
FP8 Kernels (optional)
FP8 kernels developed for LTX-Video provide performance boost on supported graphics cards (Ada architecture and later). To install FP8 kernels, follow the instructions in that repository.
Inference
📝 Note: For best results, we recommend using our ComfyUI workflow. We're working on updating the inference.py script to match the high quality and output fidelity of ComfyUI.
To use our model, please follow the inference code in inference.py:
For image-to-video generation:
python inference.py --prompt "PROMPT" --conditioning_media_paths IMAGE_PATH --conditioning_start_frames 0 --height HEIGHT --width WIDTH --num_frames NUM_FRAMES --seed SEED --pipeline_config configs/ltxv-13b-0.9.8-distilled.yaml
Extending a video:
📝 Note: Input video segments must contain a multiple of 8 frames plus 1 (e.g., 9, 17, 25, etc.), and the target frame number should be a multiple of 8.
python inference.py --prompt "PROMPT" --conditioning_media_paths VIDEO_PATH --conditioning_start_frames START_FRAME --height HEIGHT --width WIDTH --num_frames NUM_FRAMES --seed SEED --pipeline_config configs/ltxv-13b-0.9.8-distilled.yaml
For video generation with multiple conditions:
You can now generate a video conditioned on a set of images and/or short video segments.
Simply provide a list of paths to the images or video segments you want to condition on, along with their target frame numbers in the generated video. You can also specify the conditioning strength for each item (default: 1.0).
python inference.py --prompt "PROMPT" --conditioning_media_paths IMAGE_OR_VIDEO_PATH_1 IMAGE_OR_VIDEO_PATH_2 --conditioning_start_frames TARGET_FRAME_1 TARGET_FRAME_2 --height HEIGHT --width WIDTH --num_frames NUM_FRAMES --seed SEED --pipeline_config configs/ltxv-13b-0.9.8-distilled.yaml
Using as a library
from ltx_video.inference import infer, InferenceConfig
infer(
InferenceConfig(
pipeline_config="configs/ltxv-13b-0.9.8-distilled.yaml",
prompt=PROMPT,
height=HEIGHT,
width=WIDTH,
num_frames=NUM_FRAMES,
output_path="output.mp4",
)
)
ComfyUI Integration
To use our model with ComfyUI, please follow the instructions at https://github.com/Lightricks/ComfyUI-LTXVideo/.
Diffusers Integration
To use our model with the Diffusers Python library, check out the official documentation.
Diffusers also support an 8-bit version of LTX-Video, see details below
Model User Guide
📝 Prompt Engineering
When writing prompts, focus on detailed, chronological descriptions of actions and scenes. Include specific movements, appearances, camera angles, and environmental details - all in a single flowing paragraph. Start directly with the action, and keep descriptions literal and precise. Think like a cinematographer describing a shot list. Keep within 200 words. For best results, build your prompts using this structure:
- Start with main action in a single sentence
- Add specific details about movements and gestures
- Describe character/object appearances precisely
- Include background and environment details
- Specify camera angles and movements
- Describe lighting and colors
- Note any changes or sudden events
- See examples for more inspiration.
Automatic Prompt Enhancement
When using LTXVideoPipeline directly, you can enable prompt enhancement by setting
LTX-Video 是 Lightricks 公司基于 DiT 的视频生成模型的官方代码仓库。该模型可以生成同步的音频和视频，支持文生视频、图生视频、关键帧条件控制、视频延长以及视频到视频的转换。