Lightricks · 2026-03-05 · major
LTX-2.3 — open-source 4K video+audio generation
22B-parameter open-source DiT model that generates synchronized 4K video and audio in a single pass. Native portrait (9:16), rebuilt VAE for sharper output, Apache 2.0 license.
The first open-source model that generates native 4K video with synchronized audio in one forward pass — under Apache 2.0.
Key specs
| License | Apache 2.0 (<$10M revenue) |
|---|---|
| Parameters | 22B |
| Resolution | up to 4K @ 50fps |
What is it?
LTX-2.3 is a 22-billion-parameter Diffusion Transformer model from Lightricks, released March 5, 2026. It generates synchronized video and audio within a single model, supporting text-to-video, image-to-video, video-to-video, and audio-to-video workflows. It is the highest-performing open-weight video generation model available, and the only open-source model generating native 4K with audio.
How does it work?
Three core components were rebuilt from LTX-2: a new VAE that produces sharper output with better texture and facial detail, a 4x larger text connector for better prompt adherence and reduced prompt drift, and native portrait (9:16) support that generates 1080x1920 natively instead of cropping horizontal frames. The model runs locally with a companion desktop editor. Distilled and LoRA variants are available for different hardware targets.
Why does it matter?
Open-source video generation has lagged behind proprietary offerings in both quality and audio support. LTX-2.3 closes both gaps while running locally, which matters for creators who need to iterate fast without API costs and for teams that need to keep footage private. The Apache 2.0 license (for companies under $10M revenue) makes commercial use straightforward for most startups.
Who is it for?
Video creators, content teams, researchers working on generative video.
Try it
huggingface.co/Lightricks/LTX-2.3