First open-source DiT-based foundation model for synchronized 4K video and audio generation with 19B parameters
Experience AI-powered 4K video and audio generation in real-time
Explore in-depth guides and practical tutorials for LTX-2
Comprehensive guide to MOSS-TTS, the revolutionary open-source TTS model with 1.7B/0.6B variants, 10 languages, 49 voices, and ultra-low latency streaming.
Read More
The most powerful open-weight language model with 397B total parameters and 17B active per forward pass using MoE architecture.
Read More
Comprehensive guide to ACE-Step 1.5, the powerful open-source multimodal model with 32B parameters, Qwen2.5-32B backbone, and state-of-the-art vision-language performance.
Read More
Comprehensive guide to Kani-TTS-2, the revolutionary open-source TTS model with 2.5B/0.9B variants, 12 languages, 60+ voices, and ultra-low latency streaming.
Read More
A specialized image editing model offering high-fidelity results in restoration, enhancement, style transfer, and object manipulation.
Read More
Zhipu AI's latest open-source language model series with 9B+ parameters, 128K context, and multiple variants including GLM-5-Chat, GLM-5-Plus, and GLM-5-Flash.
Read More
Revolutionary multilingual speech recognition with 52 languages, 5.2% Chinese WER, 7.8% English WER, and 0.3x real-time factor.
Read More
Revolutionary OCR model achieving 91.09% accuracy with DeepEncoder V2 architecture and human-like reading patterns.
Read More
Revolutionary 6B parameter model ranked #1 among open-source image generators with single-stream architecture.
Read More
Revolutionary open-source text-to-speech model with 10 languages, 49 voices, and 3-second voice cloning.
Read More
Microsoft's breakthrough ASR model for 60-minute single-pass transcription with speaker diarization.
Read More
Groundbreaking 4B parameter agent model that matches 8B models with deep exploration capabilities.
Read More
First open-source autoregressive image model with exceptional text rendering for posters and diagrams.
Read More
Professional-grade image generation on consumer hardware with sub-second inference times.
Read More
Discover AI model that automatically decomposes images into editable RGBA layers.
Read MoreExplore the advanced features that make LTX-2 the leading open-source AI video generation model
Generate high-quality videos from text prompts with LTX-2's advanced DiT architecture
Transform static images into dynamic videos with smooth motion and natural transitions
Create perfectly synchronized audio and video content in a single unified model
Generate production-ready 4K videos with spatial upscaling capabilities
Customize LTX-2 for specific styles, motions, or appearances with efficient LoRA training
Choose from dev, distilled, or quantized (fp8/fp4) models for optimal speed-quality balance
LTX-2 leverages cutting-edge Diffusion Transformer technology with 19B parameters
LTX-2 is built on a Diffusion Transformer (DiT) architecture, the first of its kind to generate synchronized audio and video in a single unified model. With 19 billion parameters, it delivers production-ready quality for professional workflows.
Discover how LTX-2 empowers creators across industries
Generate engaging social media videos from text descriptions with LTX-2's text-to-video capabilities
Rapid prototyping and pre-visualization for filmmakers using LTX-2's 4K generation
Create promotional videos with synchronized audio using LTX-2's audio-visual synthesis
Produce educational content and tutorials with LTX-2's image-to-video animation
Experiment with AI video generation techniques using LTX-2's open-source architecture
Generate cinematic cutscenes and trailers with LTX-2's video-to-video transformation
Explore stunning examples generated by LTX-2
A dramatic sunset over mountains with flowing clouds
Static portrait brought to life with natural motion
Synchronized audio and video generation
Transform existing video with new artistic style
LTX-2 fine-tuned for specific artistic style
Spatial and temporal upscaling demonstration
Install and run LTX-2 locally in minutes
git clone https://github.com/Lightricks/LTX-2.git
cd LTX-2
uv sync
source .venv/bin/activateClone the LTX-2 repository and set up the environment using uv package manager
Python Version
β₯ 3.12
CUDA Version
> 12.7
PyTorch Version
~ 2.7
Find answers to common questions about LTX-2
LTX-2 is a 19B parameter DiT-based AI foundation model for synchronized audio-video generation. It's the first open-source model of its kind, capable of generating high-quality 4K videos with synchronized audio from text prompts, images, or existing videos.
LTX-2 supports multiple generation modes: text-to-video, image-to-video, video-to-video transformation, audio-to-video, and joint audio-visual content creation. It can generate videos up to 4K resolution with synchronized audio.
LTX-2 requires Python β₯3.12, CUDA >12.7, PyTorch ~2.7, and an NVIDIA GPU with sufficient VRAM. The exact VRAM requirements depend on the model variant and generation settings you choose.
Yes, LTX-2 is fully open-source under the Apache 2.0 license. You can freely use, modify, and distribute LTX-2 for both personal and commercial projects.
LTX-2 offers several variants: dev (bf16 full precision), fp8 and fp4 quantized versions for faster inference, and a distilled version optimized for speed. Additionally, spatial and temporal upscaler models are available.
Yes, LTX-2 supports LoRA fine-tuning for custom styles, motions, and appearances. You can train motion, style, or likeness LoRAs in less than 1 hour in many settings.
LTX-2 supports up to 4K resolution with spatial upscaling capabilities. The base model generates videos at various resolutions, and the spatial upscaler can enhance them to 4K quality.
Generation time depends on the model variant you choose. The distilled version is fastest with 8 steps, while the dev version offers highest quality but takes longer. Quantized versions (fp8/fp4) provide a good balance.
Yes, LTX-2 is the first DiT model to generate synchronized audio and video in a single model. It can create perfectly matched audio-visual content for various applications.
You can try the live demo on HuggingFace Spaces at huggingface.co/spaces/Lightricks/ltx-2-distilled, or install LTX-2 locally from GitHub for full control and customization.