LTX 2.3 AI Video Generator
Open-Source Video AI, Reimagined
Lightricks' LTX 2.3 delivers sharper details, stronger prompt adherence, and videos up to 20 seconds or 4K resolution. With native portrait support and cleaner audio, it's the most capable open-source video model yet.
Text to Video
See What LTX 2.3 Can Create
From cinematic text prompts to portrait reels, experience the next generation of open-source video AI
Prompt Enhancement
Image to Video
Portrait Video (9:16)
Cleaner Audio
Fine Detail Generation
4K Quality
Prompt Enhancement
Image to Video
Portrait Video (9:16)
Cleaner Audio
Fine Detail Generation
4K Quality
Powerful Features in LTX 2.3
Key improvements that make LTX 2.3 the most capable open-source video model available
Sharper Fine Details
Rebuilt VAE architecture trained on higher-quality data generates sharper textures, hair, and edge detail at every resolution. Reduces the need for post-processing sharpening or upscaling.
Stronger Prompt Adherence
A 4x larger text connector accurately handles complex prompts with multiple subjects, spatial relationships, and stylistic instructions - without sacrificing visual consistency.
Up to 20-Second Videos
Generate high-fidelity videos up to 20 seconds long, with support for resolutions up to 4K and frame rates of 24, 25, 48, or 50 FPS. Move from short clips to complete narrative sequences in a single generation.
Native Portrait Video
Supports vertical resolutions up to 2160×3840, trained on real portrait data rather than cropped landscape content. Perfect for TikTok, Instagram Reels, and YouTube Shorts.
Pro Flow & Fast Flow
Choose between Pro Flow for maximum visual quality and consistency, or Fast Flow for rapid iteration and lower compute. Both modes support text-to-video and image-to-video.
Cleaner Audio Generation
Filtered training data and a new vocoder implementation reduce audio artifacts and alignment issues. Applies to both text-to-video and audio-conditioned video workflows.
LTX 2.3 vs LTX 2
See what's new and improved in Lightricks' latest open-source video model
| Capability | LTX 2.3New | LTX 2 |
|---|---|---|
| Native Portrait (9:16) | ||
| Detail Quality | Rebuilt VAE | Standard |
| Prompt Adherence | 4× Larger Connector | Standard |
| Image-to-Video Motion | Natural Motion | Some Freezing |
| Audio Quality | Cleaner (New Vocoder) | Standard |
| Max Resolution | 4K | 4K |
| Max Duration | 20 seconds | 20 seconds |
What Creators Say About LTX 2.3
Join creators worldwide generating stunning open-source AI videos with LTX 2.3
The detail quality in LTX 2.3 is on another level. Hair, fabric textures, fine edges - it handles everything I throw at it without needing extra upscaling. It's become my go-to model for client work.
Portrait video support finally works the way it should. I create vertical content for Instagram Reels and TikTok daily, and LTX 2.3 generates vertical compositions naturally - no more awkward crops from landscape videos.
I can actually trust my prompts now. Complex scene descriptions with multiple subjects and spatial details used to get jumbled. With LTX 2.3 the output matches exactly what I described. Huge time saver.
The 20-second generation limit opens up storytelling possibilities that weren't there before. I can create complete product demo sequences without stitching clips together. My conversion rates have jumped since switching.
Frequently Asked Questions
Everything you need to know about LTX 2.3 AI video generation
LTX 2.3 is Lightricks' latest open-source video generation model released on March 8, 2026. It brings key improvements over LTX 2: a rebuilt VAE for sharper details, a 4x larger text connector for stronger prompt adherence, improved image-to-video quality, cleaner audio generation, native portrait video support up to 2160×3840, and Pro Flow & Fast Flow generation modes.
LTX 2.3 offers two generation flows. Pro Flow prioritizes maximum visual quality and consistency - ideal for final renders and professional content. Fast Flow optimizes for speed and rapid iteration with lower compute cost, making it perfect for quick previews and experimenting with prompts.
Yes! LTX 2.3 natively supports vertical resolutions up to 2160×3840. Unlike earlier models that crop landscape footage, LTX 2.3 was trained on real portrait-oriented data, producing natural vertical compositions ideal for TikTok, Instagram Reels, YouTube Shorts, and mobile-first content.
LTX 2.3 features a 4x larger text connector compared to LTX 2. This allows it to accurately interpret prompts involving multiple subjects, complex spatial relationships, and specific stylistic instructions. You can now describe detailed scenes with confidence and get results that closely match your vision.
Yes, all videos you create with LTX 2.3 through Veevid are yours to use for personal and commercial purposes - including social media content, advertising, client projects, and more - without additional licensing fees.
Discover More AI Tools
Explore our full suite of AI-powered video and image creation tools
Ready to Create with LTX 2.3?
Experience the most capable open-source video AI. Generate videos up to 20 seconds or 4K resolution with native portrait support and cleaner audio.
Get Started Now