LTX-2.3
Next-Gen AI Video Generation
Generate cinematic AI videos with LTX-2.3 — the open-source 22-billion-parameter diffusion transformer model delivering 4K output, native audio sync, and multi-modal input.
LTX-2.3 AI Video Generator
What Makes LTX-2.3 Stand Out?
LTX-2.3 is an advanced open-source video model built on DiT architecture that delivers sharper detail and better prompt adherence than its predecessors.
Rebuilt VAE for Sharper Video Output
The LTX-2.3 model features a completely rebuilt VAE trained on higher-quality data, preserving fine textures, hair strands, readable text, and crisp edge detail across every generated frame.
4x Larger Text Connector for Complex Prompts
With a text connector four times larger than previous versions, LTX video 2.3 handles multi-subject scenes, spatial relationships, and detailed stylistic instructions with remarkable accuracy.
Key Features of LTX-2.3
From text-to-video to native portrait generation, LTX-2.3 provides a complete multi-modal AI creation pipeline.
Text-to-Video Generation
Describe any scene in natural language and LTX-2.3 generates cinematic output with fluid motion, accurate lighting, and realistic physics simulation.
Image-to-Video Conversion
Upload a still image and the LTXV 2.3 model transforms it into a dynamic clip with natural camera motion and lifelike animation — less freezing, fewer artifacts.
Audio-to-Video Synchronization
Feed audio tracks into the LTX-2.3 pipeline and generate visuals that match beats, dialogue timing, and environmental soundscapes with precision.
Native Portrait Video (9:16)
LTX-2.3 creates vertical video natively at 1080x1920 resolution — trained on real portrait data, not cropped from landscape — perfect for TikTok, Reels, and Shorts.
Up to 4K at 50 FPS
LTX-2.3 outputs at resolutions up to 4K with frame rates of 24, 25, 48, or 50 fps, supporting professional broadcast and cinematic workflows.
Open Source and Free to Use
LTX-2.3 model weights are freely available on Hugging Face — use them for personal and commercial projects with no cost for organizations under $10M annual revenue.
Use Cases for LTX-2.3
From social media content to film pre-production, LTX-2.3 powers diverse professional creative workflows.
Social Media Content Creation
Generate consistent short-form content for TikTok, Instagram Reels, and YouTube Shorts using LTX-2.3 — produce variations for A/B testing without reshooting, cutting production time from days to hours.
Marketing and E-commerce Video
Produce product demonstrations with brand-consistent visuals using LTX-2.3. Upload reference images to maintain logos, packaging, and color grading across all marketing materials at scale.
Film Pre-Production and Previsualization
Test camera angles, visual effects, and lighting setups with LTX-2.3 before live-action shoots. Create pitch-ready cinematic sequences from storyboard sketches efficiently.
Education and Training Content
Transform lesson plans and training materials into engaging visual content using LTX-2.3. Illustrate complex concepts with AI-generated explanations and animated demonstrations.
How to Use LTX-2.3
Create cinematic AI content with LTX-2.3 in three simple steps — no editing skills required.
Enter Your Prompt
Describe the video you want in natural language, or upload an image or video clip as a creative reference for LTX-2.3.
Choose Settings
Set video duration, aspect ratio, and output quality to match your target platform and creative vision.
Generate and Download
Click generate, wait a moment, and download your high-definition AI video — ready to publish or edit further.
How LTX-2.3 Compares to Other Models
A comparison of key capabilities across leading AI generation models and the LTX-2.3 open-source platform.
LTX-2.3 vs Sora 2
Sora 2 focuses on physics simulation with up to 20-second clips. LTX-2.3 differentiates with its open-source availability, multi-modal generation pipeline, and native audio synchronization — offering comparable quality without subscription lock-in.
LTX-2.3 vs Veo 3.1
Veo 3.1 targets 4K broadcast-ready output. The LTXV 2.3 model matches with native 4K at 50 fps while offering full open-source access, LoRA microtuning support, and flexible local or cloud deployment options.
LTX-2.3 vs Kling 3.0
Kling 3.0 excels at natural motion with affordable pricing. LTX video 2.3 offers broader resolution support up to 4K, native portrait format, and open-source weights — letting developers self-host or customize with LoRA fine-tuning.
Why Choose the LTX Video Model
As one of the fastest open-source video generators available, the LTX diffusion transformer runs 18x faster than comparable models on H100 GPUs. Combined with multi-modal input and professional output quality, it delivers studio-grade results accessible to everyone.
Frequently Asked Questions
Everything you need to know about LTX-2.3 — capabilities, access, and technical details.
LTX-2.3 is the latest open-source AI video generation model from Lightricks, built on Diffusion Transformer (DiT) architecture with approximately 22 billion parameters. It generates high-fidelity video and synchronized audio from text prompts, images, or audio input — supporting resolutions up to 4K at 50 fps with durations up to 20 seconds.
LTX-2.3 introduces four major upgrades: a rebuilt VAE for sharper textures and edge detail, a 4x larger text connector for better prompt adherence with complex multi-subject scenes, stronger image-to-video generation with less freezing and fewer artifacts, and cleaner audio output with reduced drops and improved synchronization.
The LTX video model supports text-to-video, image-to-video, audio-to-video, and video-to-video generation. You can describe scenes in natural language, upload reference images for animation, provide audio tracks for synchronized visuals, or transform existing video clips with precise creative control.
LTX-2.3 outputs at 1080p, 1440p, and 4K resolutions with frame rate options of 24, 25, 48, or 50 fps. It supports both landscape and native portrait (9:16) aspect ratios — trained on real vertical video data rather than cropped from landscape footage.
Yes. The LTX-2.3 model weights are open-source on Hugging Face under the LTX Model License, free for individuals and companies with annual revenue under $10 million. Commercial licenses are available for larger organizations. You can also use it through this platform with free credits for new users.
Yes. Lightricks provides full dev checkpoints in bf16, quantized fp8 variants for smaller VRAM footprints, and distilled models for faster inference. ComfyUI custom nodes, reference workflows, and the LTX Desktop application let you run the LTX diffusion transformer entirely on your local hardware.
Yes. The LTX open-source video model features an integrated audio generation subnetwork that produces synchronized environment sounds, effects, dialogue, and music aligned with the visual content. Audio quality has been improved with filtered training data and new vocoder technology.
LTX-2.3 is optimized for speed, running at 50 fps generation and performing 18x faster than WAN 2.2 on H100 GPUs. Two generation modes are available: Fast for rapid iteration and Pro for maximum visual quality — both significantly outperforming comparable open-source models in throughput.
Yes. The model supports LoRA microtuning for personalization and custom style adaptation. Note that custom LoRAs trained for previous LTX versions will need to be retrained for the LTX-2.3 latent space before migrating to the new model.
You can use LTX-2.3 directly through this platform with free credits for new users. The model is also available via the LTX API for managed endpoints, on Hugging Face for self-hosting, through ComfyUI for node-based workflows, and via the LTX Desktop application for fully local generation.
Start Creating AI Videos with LTX-2.3
Generate professional cinematic content from text, images, or audio — free credits for new users on the LTX-2.3 platform.