Highlights
What you can do with it
- Generate short clips with matching audio and video
- Fine tune the model for a specific style or domain
- Research synchronized audiovisual generation
How to use it
- 1 Download the model weights and inference pipeline
- 2 Set up the environment on your own hardware
- 3 Run text to video or image to video generation
- 4 Fine tune under the Apache 2.0 license if needed
Who it is for
Pros
- Fully open release with weights and training code
- 4K at up to 50 frames per second
- Apache 2.0 license allows customization
Limitations
- Self hosting a 19B model needs serious GPU resources
- Clips are limited to 20 seconds
- Open source releases can change licensing or support over time
What ltx is
LTX-2 is an AI foundational model released by Lightricks in January 2026 for generating synchronized audiovisual content. Its core advantage, as the page puts it, is being the first open source model capable of generating synchronized audio and video, meaning the sound and picture are produced together rather than stitched in afterward. The model supports 4K resolution at up to 50 frames per second and can produce video clips up to 20 seconds long. Under the hood it is based on a 19 billion parameter DiT architecture, with what the team describes as strong optimizations in both speed and quality. The release is not only free but also provides complete model weights, training code, and inference pipelines, so developers can fine tune, customize, and deploy it under the Apache 2.0 license. For researchers, having the training code matters as much as the weights, because it makes the work reproducible rather than just usable.
What you can do with it
- Generate short clips with synchronized audio and video
- Fine tune the model for a specific style or domain
- Deploy on your own infrastructure under Apache 2.0
- Use it for content creation, film and animation, training, and research
- Study the training code to reproduce or extend the work
Who it is for
- Developers who want open weights for audiovisual generation
- Researchers studying synchronized audio and video models
- Studios and educators exploring film, animation, and training content
- Developers fine tuning open models for audio and video
What to watch out for
A 19 billion parameter model is not a laptop project: self hosting needs serious GPU memory and engineering time, and inference at 4K and 50 frames per second demands strong hardware. Clips are limited to 20 seconds, so longer sequences still require editing and assembly around the model. Open source projects can change licensing, weights, or support over time, so pin the version you deploy. As with any generative model, review outputs for artifacts before they go into anything public. Plan storage as carefully as compute, because 4K clips at 50 frames per second are large, and a project can fill a disk faster than expected.
FAQ
Who released LTX-2?
Lightricks released it in January 2026.
What makes it notable?
It is described as the first open source model capable of generating synchronized audio and video.
What are the specifications?
4K resolution at up to 50 frames per second, clips up to 20 seconds, and a 19 billion parameter DiT architecture.
Can I deploy it myself?
Yes, it is free with full weights, training code, and inference pipelines under Apache 2.0.
Why it is free
What the official page says： 免费
Free limits, trial length and features are decided by the official site and can change at any time.
Product preview
Screenshot archived from the listing page, for preview only.