Lightricks, the app developer behind LTX Studio, Facetune, and Videoleap, has released its first custom AI video model: LTX Video 0.9. LTX Video 0.9 is open source and can generate a 5-second AI video in as little as 4 seconds.

According to the company, this new model can generate clips with impressive motion consistency and realism, and does so more efficiently than other comparable alternatives.

LTX Video is called a “real-time video creator” built on feedback from users of LTX Studio, a platform that allows users to create multi-clip projects from a single prompt.

The 4-second, 5-second video was generated using an Nvidia H100 GPU at a resolution of 768 x 512. This model will also work with a standard Nvidia RTX 4090, but in that case the video would take considerably longer than 4 seconds to produce.

Zeev Farbman, co-founder and CEO of Lightricks, said: “With many AI technologies becoming proprietary, we can help the global academic and developer community build and shape the future of AI video, We believe it is time for an open source video model.”

LTX Video, like Mochi-1, is open source and, according to Lightricks, can quickly generate video while maintaining video quality and motion accuracy.

“We built Lightricks with the vision of pushing the boundaries of what is possible in digital creativity to continue to bridge the gap between imagination and creation. Ultimately leading to LTXV, we will be able to develop better products that address the needs of so many industries that leverage the power of AI,” Farbman said.

Being able to comfortably run such models on a good gaming PC is a big step up for AI video, allowing it to be integrated into gaming and video editing tools for real-time rendering and previewing. [The company promises “unparalleled motion and structure consistency” with the LTX Video diffusion transformer architecture. This ensures consistent transitions between individual frames in a 5-second video, creating smoother motion and reducing morphing.

According to Lightricks CTO Yaron Inger, this will facilitate scaling up to longer video productions in the future. He states that a wider range of use cases will be possible than what is possible today

: “The ability to generate video faster than it can be played back opens the possibility for applications beyond content creation, such as gaming, interactive experiences for shopping, learning, and socializing We believe that researchers and developers will be able to use this technology to create a new type of video experience. We look forward to seeing how researchers and developers build on this underlying model.”

We tried LTX Video in an early preview and were impressed not only with the overall visual output, but also with the quality of the motion; while not as good as Kling or Runway Gen-3, it is a serious contender for an open source model that can be generated at speed.

It has image-to-video and text-to-video modes. LTX Video is also useful in production scenarios because it can accommodate different video lengths and resolutions.

Finally, the code base and model weights are completely open source, allowing other developers to extend and build on the base model. We have experienced this with image models such as Flux and Stable Diffusion, which result in a wider range of functionality than any one company could develop alone.

All of the videos in this article were created using LTX Video and were generated as fast as expected. We have not yet tried it offline, but it is available from ComfyUI if you have a sufficiently powerful gaming PC.

If you do not have a gaming PC, it is available from HuggingFace and FAL.ai.