Recently, Israeli tech company Lightricks released the latest version of its open-source AI video generation model — LTX-Video-13B distilled model (LTXV-13B v0.9.7-distilled). Built on a 13 billion parameter foundation, this model uses multi-scale rendering technology and efficient quantization optimization to accelerate video generation to an impressive speed of less than 10 seconds, marking a new milestone in AI video generation technology. AIbase delves into this breakthrough advancement, guiding you through its technical highlights and industry impact.
Project address: https://github.com/Lightricks/LTX-Video
Model address: https://huggingface.co/spaces/Lightricks/ltx-video-distilled
Technical Breakthrough: HD Video Generation in 10 Seconds
The LTX-Video-13B distilled model adopts advanced multi-scale rendering technology, gradually refining low-resolution sketches to high-definition quality, significantly shortening video generation time. According to AIbase's understanding, this model can generate videos at a resolution of 1216×704 on consumer-grade GPUs like NVIDIA RTX4090 within 10 seconds, speeding up by nearly five times compared to the previous generation model. Low-resolution previews can even take only 3 seconds. This speed breakthrough makes real-time video generation possible, providing creators with unprecedented efficiency.
In addition, the model supports keyframe editing, camera motion control, and multi-shot sequences, giving users greater creative freedom. Whether generating dynamic scenes or complex character movements, LTXV-13B can present smoother motion trajectories and clearer details, with video quality comparable to professional film works.
Open Source Empowerment: High-Quality Video Generation on Low-Memory Devices
As a fully open-source model, the LTX-Video-13B distilled model continues Lightricks' commitment to technological inclusiveness. AIbase noticed that the model particularly offers quantized versions (fp8) and LoRA-compatible versions, greatly lowering hardware requirements. Even on low-memory devices, users can run the model to generate high-quality videos. This feature is especially suitable for independent creators and small-to-medium enterprises, significantly reducing the cost of AI video production.
The model is available for download on Hugging Face and GitHub, and it provides compatibility support through ComfyUI, allowing developers to easily integrate it into existing workflows. Lightricks also offers free licenses to businesses with annual revenues below $10 million, further promoting the popularity of AI video technology.
Efficiency and Quality Combined: The Secret Behind 30x Speed Improvement
The LTX-Video-13B distilled model not only achieves remarkable speed but also significantly enhances video quality. Thanks to collaborations with Shutterstock and Getty Images, the model has been trained on high-quality video datasets, resulting in more cinematic visuals and stronger scene continuity. AIbase learned that its core technology, UEfficient Q8 kernel optimization, improves the model's performance on consumer-grade hardware, ensuring efficient operation even on ordinary laptops.
Compared to similar models, LTXV-13B generates videos 30 times faster, with an average generation time of just 37.59 seconds, while competitors on equivalent hardware may require nearly 25 minutes. This leap in efficiency gives it great potential in real-time generation, virtual influencers (VTubing), and other scenarios.
Industry Impact: A Challenge to OpenAI and Google from Native Forces
Lightricks' release is seen by the industry as a strong challenge to major models like OpenAI Sora and Google Veo. AIbase analysis suggests that the LTX-Video-13B distilled model achieves top-tier performance at a lower training cost (only tens of millions of dollars), showcasing the innovation capabilities of open-source communities and smaller tech companies in the AI field. Its contributions of multi-scale rendering and VACE model inference have injected new vitality into global AI video generation technology development.
For creators, the arrival of LTXV-13B means they no longer need expensive professional equipment to produce high-quality animations, short films, or marketing videos. AIbase predicts that this model will accelerate the popularization of AI video tools in the film, advertising, and social media sectors, reshaping the content creation ecosystem.
The "Deepseek Moment" in AI Video Generation
As a professional AI media outlet, AIbase believes that the release of the LTX-Video-13B distilled model is not only Lightricks' peak technical achievement but also a collective victory for the open-source AI community. Its ultra-fast generation capability and low hardware requirements break down the technical barriers in AI video generation, enabling more people to participate in intelligent creation.