Lightricks Unveils LTX-Video-13B: AI Video Generation in 10 Seconds
Israeli technology firm Lightricks has introduced its latest innovation in AI video generation—the LTX-Video-13B distilled model (LTXV-13B v0.9.7-distilled). This open-source model, built on a 13-billion-parameter foundation, leverages multi-scale rendering and quantization optimization to achieve unprecedented generation speeds of under 10 seconds, setting a new benchmark for the industry.
Technical Advancements: Speed Meets Quality
The LTX-Video-13B model employs multi-scale rendering to refine low-resolution sketches into high-definition videos efficiently. On consumer-grade GPUs like the NVIDIA RTX4090, it generates 1216×704 resolution videos in just 10 seconds—a fivefold improvement over its predecessor. Low-resolution previews take as little as 3 seconds, enabling near-real-time creation. The model also supports advanced features like keyframe editing, camera motion control, and multi-shot sequences, offering filmmakers and content creators unparalleled flexibility.
Open-Source Accessibility
Lightricks has prioritized inclusivity by releasing the model as fully open-source, with quantized (fp8) and LoRA-compatible versions available. These optimizations allow the model to run efficiently on low-memory devices, democratizing access to high-quality AI video tools. The model is downloadable via Hugging Face and GitHub, with free licenses offered to businesses earning under $10 million annually.
Efficiency Breakthrough
The LTXV-13B model isn’t just fast—it’s also remarkably efficient. Collaborations with Shutterstock and Getty Images provided high-quality training data, resulting in cinematic visuals and smooth scene transitions. Its UEfficient Q8 kernel optimization ensures robust performance even on standard laptops. Compared to competitors, LTXV-13B is 30 times faster, completing tasks in 37.59 seconds versus rivals’ 25-minute averages.
Industry Implications
The release challenges giants like OpenAI’s Sora and Google’s Veo by delivering top-tier performance at a fraction of the cost (training budgets under $100 million). Analysts see this as a win for open-source innovation, with potential applications in filmmaking, advertising, and social media content creation.
Key Points
- Generates HD videos in under 10 seconds using multi-scale rendering.
- Open-source model with quantized versions for low-memory devices.
- Trained on premium datasets for cinematic quality.
- Outperforms competitors with 30x faster generation speeds.
- Free licenses available for small businesses.