“LTX Video 0.9” is a new approach that can create five-second video clips in just four seconds, with longer segment capability, superior motion consistency and efficiency
Lightricks, the company behind AI-driven storytelling platform LTX Studio and the globally successful portrait editor Facetune, announced today the release of its open-source video generation model, called LTX Video (LTXV 0.9). With a breakthrough capability to generate five (5) seconds of high-quality video in only four (4) seconds, LTXV sets a new standard for real-time AI video creation. Developed with direct feedback from LTX Studio users, LTXV is a groundbreaking evolution in AI video generation, providing a new framework for researchers to further advance solutions for motion consistency, computational efficiency, and scalability while setting new standards in performance.
"We built Lightricks with a vision to push the boundaries of what's possible in digital creativity to continue bridging that gap between imagination and creation - ultimately leading to LTXV, which will allow us to develop better products that address the needs of so many industries taking advantage of AI’s power," said Zeev Farbman, Co-founder and CEO of Lightricks. “With many AI technologies becoming proprietary, we believe it’s time for an open-sourced video model that the global academic and developer community can build on and help shape the future of AI video.”
Shaping the Future of AI Video Technology
This new model represents a significant leap in generative AI video capabilities, addressing key industry pain points while setting new performance benchmarks. Running on NVIDIA’s H100 GPUs, LTXV achieves remarkable real-time processing speeds: generating a five-second video sequence (121 frames at 768 x 512 resolution) in just four (4) seconds using 20 diffusion steps. This throughput surpasses current generation speeds of most existing models while maintaining top-tier visual quality.
At two billion parameters, LTXV delivers enterprise-grade performance on accessible hardware. Unlike many AI models that rely on costly high-end GPUs or require aggressive quantization (which can sacrifice quality to save memory), LTXV is designed to maintain precision and visual quality without compromising speed or memory efficiency. Running on bfloat16 precision (which provides nearly the same range as full precision but uses less memory), LTXV efficiently generates high-quality video even on widely available prosumer GPUs, such as the RTX 4090. This makes LTXV particularly suited for smaller creative studios and independent creators who want to get immediate results and feedback and iterate effectively, bringing cutting-edge video generation within reach on commonly available configurations.
The initial release of Lightricks’ LTXV includes:
- Unmatched motion and structure consistency: LTXV’s Diffusion Transformer architecture ensures smooth, coherent transitions between frames, eliminating issues like object morphing that have plagued previous generations of video models. Diffusion models are particularly good at creating high-quality samples, which makes them strong tools for things such as synthesizing images.
- Rapid processing and rendering times: Optimized for both GPU and TPU systems, LTXV reduces video generation times by more than 90%, making it among the fastest model of its kind available for high-quality video output.
- Scalable long-form video production: Capable of producing extended, high-quality videos with consistency across length, offering creators more flexibility and control.
“LTXV represents a new era of AI-generated video,” said Yaron Inger, CTO of Lightricks. “By designing a robust video encoding model that compresses the video in a very compact way (1:192 of the original video), we’ve achieved unprecedented speed while improving motion consistency and visual continuity. The ability to generate videos faster than playing them opens the possibility for applications beyond content creation, like gaming and interactive experiences for shopping, learning or socializing. We're excited to see how researchers and developers will build upon this foundational model.”
Impact on LTX Studio
The integration of LTXV into LTX Studio significantly enhances the platform, allowing creators to generate longer, more dynamic videos with greater speed and precision. This advancement enables users to focus on creativity rather than technical barriers, setting a new standard in AI-powered video production.
Empowering Global Innovation with Open-Source Contributions
Reinforcing Lightricks’ commitment to the collaborative process, LTXV follows in the footsteps of Long AnimateDiff, an open-source animation framework released earlier this year by Lightricks. LTXV is designed for extensive customization, allowing developers to fine-tune outputs to meet specific needs, fostering a global community of experimentation and advancement.
Released on Github and Hugging Face, the model was built using PyTorch XLA and runs seamlessly on both GPU and TPU systems.
Efficiency and Future Roadmap
In terms of computational efficiency, LTXV’s architecture allows it to achieve high-quality results comparable to larger models but without the heavy computational load (at only 2-billion parameters.) Lightricks plans future updates focused on enhanced motion consistency, higher resolutions, and more advanced customization features, while continually reducing compute times. For more information about Lightricks and its open-source initiatives, please visit [www.lightricks.com].
.png)
