Introducing LTX-2.3 Our most production-ready model yet. The fastest 4K video generation in the world with built-in native dialogue. Here’s what’s new 🧵 1/9
LTX-2.3 AI Video Generator
Lightricks LTX-2.3 is a 22B parameter DiT-based audio-video foundation model with a rebuilt VAE for sharper textures, faces, hair, and text rendering. Features a 4x larger text connector for superior prompt adherence, native 9:16 portrait mode up to 1080x1920, LoRA support with up to 3 custom adapters, and cleaner audio via HiFi-GAN vocoder. Generates 480p to 1080p videos from 5 to 20 seconds with synchronized audio.
/modelsLightricks Ltx
Reel · Specifications
What is LTX-2.3?
Lightricks' 22B parameter DiT-based audio-video foundation model with rebuilt VAE and open-source weights
- · 0122BParameters
- · 02Up to 4KResolution
- · 035-20sDuration
- · 04Apache 2.0License
A next-generation open-source model delivering sharper details, better prompt adherence, native portrait video, and cleaner synchronized audio under Apache 2.0 license.
Reel · Capabilities
LTX-2.3 Features
Discover the powerful capabilities of Lightricks LTX-2.3 for next-generation video production
- Feature 01 / 08
Rebuilt VAE Engine
Completely rebuilt Variational Autoencoder delivers dramatically sharper fine details including textures, faces, hair strands, text overlays, and crisp edges. A fundamental upgrade over LTX-2 for professional-quality output.
- Feature 02 / 08
Enhanced Prompt Adherence
A 4x larger text connector ensures LTX-2.3 follows your prompts with significantly greater accuracy. Complex scene descriptions, specific actions, and detailed visual instructions are faithfully rendered in the generated video.
- Feature 03 / 08
Native Portrait Video
First-class 9:16 vertical video support at up to 1080x1920 resolution, purpose-built for TikTok, Instagram Reels, and YouTube Shorts. No more cropping or letterboxing from landscape-only models.
- Feature 04 / 08
Synchronized Audio-Video
Cleaner, higher-fidelity audio generation powered by a new HiFi-GAN vocoder. Produces contextually appropriate sound effects, ambient audio, and dialogue that naturally syncs with the visual content.
- Feature 05 / 08
Custom LoRA Adapters
Apply up to 3 simultaneous LoRA adapters to customize style, character consistency, or motion patterns. Fine-tune outputs for brand-specific aesthetics or specialized creative workflows.
- Feature 06 / 08
Multi-Resolution Output
Flexible resolution from 480p for rapid drafts up to 1080p for production-ready output, with upscaling support to 4K. Choose 24 or 48 FPS frame rates to match your project requirements.
- Feature 07 / 08
Improved Image-to-Video
Significantly reduced Ken Burns effect and static freezing artifacts. Images are animated with genuine motion and dynamic camera work, producing natural movement that brings still images to life.
- Feature 08 / 08
Open-Source Foundation
Released under Apache 2.0 license with full model weights and training code available. Build custom applications, fine-tune for specific use cases, or integrate directly into your production pipeline.
How to Use LTX-2.3 for Image-to-Video
Transform still images into dynamic videos with improved motion in three simple steps
Select LTX-2.3 Image-to-Video Model and Upload Image
Navigate to the video generation page and select the LTX-2.3 image-to-video model. Upload a high-quality still image (JPG, PNG, or WEBP format). Choose your resolution (480p to 1080p), aspect ratio (16:9 or 9:16 portrait), frame rate (24 or 48 FPS), and duration (5-20 seconds). Optionally attach LoRA adapters for custom styles.
Add Motion Prompt and Configure Settings
Write a detailed prompt describing how you want your image to come to life. LTX-2.3's improved image-to-video produces genuine motion with less Ken Burns effect and less freezing. Specify camera movements, subject actions, and environmental effects. Enable audio generation for synchronized HiFi-GAN sound, and use prompt optimization for best results.
Generate and Download Your Video
Click generate and wait for LTX-2.3 to animate your image with the rebuilt VAE for sharper details and natural movement. The model preserves your image's composition while adding dynamic motion and depth. Preview the result with synchronized audio, then download your finished video in MP4 format.
How to Use LTX-2.3 for Text-to-Video
Create stunning AI videos with synchronized audio in three simple steps
Select LTX-2.3 Text-to-Video Model
Navigate to the video generation page and select the LTX-2.3 text-to-video model. Choose your preferred resolution (480p to 1080p), aspect ratio (16:9 landscape or 9:16 portrait), frame rate (24 or 48 FPS), and video duration (5-20 seconds). Optionally attach up to 3 LoRA adapters for custom styles.
Write Your Prompt and Configure Settings
Craft a detailed text description of the video you want to create. LTX-2.3's 4x larger text connector ensures superior prompt adherence, so be specific about scene composition, camera movements, lighting, and actions. Enable prompt optimization for better results, toggle synchronized audio generation, and adjust any additional parameters.
Generate and Download Your Video
Click generate and wait for LTX-2.3 to render your video with the rebuilt VAE for sharper textures, faces, and text. Preview the result with synchronized HiFi-GAN audio. Download your finished video in MP4 format for immediate use in your projects, social media, or production workflows.
LTX-2.3 YouTube Videos
Watch demonstrations and tutorials of LTX-2.3, the 22B parameter open-source audio-video foundation model
- Introducing LTX Desktop: An Open Source Video Editor Powered by LTX-2.3 - LTX
- Run LTX 2.3 Video Generation AI Model Locally with ComfyUI - Easy Guide - Fahd Mirza
- LTX 2.3 in Comfy UI — Text to Video & Image to Video - AI Ninja
- LTX-2.3 In ComfyUI = AI Video Generation At 0 Credits Per Run - Nerdy Rodent
- LTX-2.3 ComfyUI Workflow Tutorial | Text-to-Video, Image-to-Video, Talking Avatar & Audio Generation - Vantage with AI
- LTX 2.3 Released - ComfyUI Workflow & A New Tool I Built To Run AI😃😃😃 - Benji’s AI Playground
LTX-2.3 YouTube Videos
Watch demonstrations and tutorials of LTX-2.3, the 22B parameter open-source audio-video foundation model
LTX-2.3 Popular Reviews on X
See what the AI community is saying about LTX-2.3, the 22B parameter open-source audio-video foundation model
Keyframes and structured control are now more deeply integrated. LTX-2.3 is trained with multi-task objectives from the pretraining stage, including image-to-video, retake, keyframes, and more. This makes transitions, controlled scene evolution, and multi-shot workflows more Show more
🚀 Today we’re releasing LTX-2.3 with open weights + training code, alongside the API, LTX Studio, and LTX-Desktop - a full-featured video editing app that runs on your local GPU. Audio+video generation just leveled up: quality + capabilities + tooling - all open-source. 🧵👇
https://t.co/MGZ9rXyOsh
Acaban de liberar LTX-2.3. Un modelo de video con IA que genera 4K + audio + lip-sync. 100% gratis y open-source. Puedes crear clips de hasta 20s desde tu propio PC. Te dejo el enlace en el comentario: (Si no, prueba lo último de Kling) x.com/ivnways/status…
Kling 3.0 Motion Control se actualiza por completo. Basado en la versión 2.6, ahora ofrece: - Consistencia facial impecable - Estabilidad desde múltiples ángulos - Eidelidad en secuencias largas - Reproducción fiel de emociones complejas
LTX 2.3 seems to be coming out soon. No models on Hugging Faces just yet, but soon I'm sure ~ ltx.io/model/ltx-2-3
LTX-2.3 releasing soon 😊 LTX-2.3 brings four major improvements over LTX-2. A redesigned VAE produces sharper fine details, more realistic textures, and cleaner edges. A new gated attention text connector means prompts are followed more closely – descriptions of timing, Show more
LTX-2.3 officially released! Wait, unreleased. 404. ComfyUI team: “Nah, we already support it.” github.com/Comfy-Org/Comf…
LTX-2.3 is a 22 billion parameter DiT-based audio-video foundation model by Lightricks, released on March 5, 2026. It features a completely rebuilt VAE for sharper textures, faces, hair, and text rendering, a 4x larger text connector for better prompt adherence, native 9:16 portrait support, LoRA adapter support, and a new HiFi-GAN vocoder for cleaner audio. It is open-source under Apache 2.0.
LTX-2.3 introduces several major upgrades: the VAE has been completely rebuilt for dramatically sharper fine details. The text connector is 4x larger, resulting in much better prompt adherence. Image-to-video generation has been improved with less Ken Burns effect and less freezing. Audio quality is cleaner thanks to a new HiFi-GAN vocoder. It also adds native 9:16 portrait video support and LoRA adapter support for customization.
LTX-2.3 generates videos from 480p up to 1080p natively, with upscaling support to 4K. It supports both landscape (16:9) and portrait (9:16) aspect ratios, with portrait mode supporting up to 1080x1920 resolution. Video duration ranges from 5 to 20 seconds, with frame rate options of 24 or 48 FPS.
LTX-2.3 supports up to 3 simultaneous LoRA (Low-Rank Adaptation) adapters that can be applied to customize the model's output. LoRAs can be used to maintain consistent character appearances, apply specific artistic styles, or control motion patterns. This allows for brand-specific customization and specialized creative workflows without retraining the full model.
LTX-2.3 generates synchronized audio-video content using a new HiFi-GAN vocoder that produces cleaner, higher-fidelity audio compared to LTX-2. The model generates contextually appropriate sound effects, ambient audio, and dialogue that naturally matches the visual content. Audio generation can be toggled on or off depending on your needs.
Yes, LTX-2.3 is released under the Apache 2.0 license, which is one of the most permissive open-source licenses available. The full model weights and training code are publicly accessible, allowing developers to build custom applications, fine-tune for specific use cases, or integrate the model directly into production pipelines for both personal and commercial use.