Lightricks announces a major update to its AI video generation model LTX-Video (LTXV), now supporting the generation of image-to-video content up to 60 seconds. This breakthrough breaks the industry standard 8-second limit, making LTXV the first open-source model capable of real-time streaming of long-form AI videos.

image.png

60-Second Video Generation: A Leap from Short Films to Long Narratives

The latest version of LTXV introduces a autoregressive streaming architecture, enabling the generation of high-quality videos up to 60 seconds from a single image. Unlike traditional models that produce short outputs, LTXV can stream video in real time during the generation process, with the first second of content appearing almost instantly, followed by continuous scene construction. This technology not only supports longer video generation but also ensures smooth and coherent motion and narrative, offering creators unprecedented space for storytelling expression.

Users can generate videos with resolutions of 768x512 or higher (such as 1216x704) and a frame rate of 24FPS by inputting detailed text prompts combined with an initial image. LTXV's multi-scale rendering technology first captures rough movements at low resolution and then gradually refines details, ensuring the best balance between speed and quality.

Real-Time Control and Creative Flexibility

The update to the LTXV model introduces a dynamic scene control feature, allowing users to adjust elements such as posture, depth, or style in real time during video generation. With Lightricks' IC-LoRA technology, creators can continuously apply control signals (such as posture detection or edge detection), achieving precise control over video details. This "direct while generating" capability transforms LTXV from a simple prompt-based tool into a true long-form narrative platform.

In addition, LTXV supports multiple generation modes, including text-to-video, image-to-video, keyframe animation, and video expansion. Users can even combine multiple images or short video clips as conditions to generate complex video content. This flexibility makes it suitable for a wide range of applications, from social media short videos to interactive educational content, and even real-time AR visual effects.

Efficiency and Open Source: Empowering a Wide Range of Creators

Another highlight of LTXV is its efficiency and hardware-friendliness. Thanks to its 13-billion parameter model architecture and bfloat16 precision optimization, LTXV can run on consumer-grade GPUs (such as NVIDIA RTX 4090 or 5090) with minimal VRAM requirements (as low as 8GB), generating a 60-second video in just a few seconds. For example, on an H100 GPU, LTXV can generate a 5-second video in 4 seconds, faster than real-time playback. In contrast, competing models like OpenAI's Sora typically require multi-GPU clusters, resulting in high computational costs.

As an open-source model, the code and weights of LTXV are freely available on GitHub and Hugging Face, under the LTXV Open Weights License, supporting academic research and community development. Lightricks has also released配套 tools, such as LTX-Video-Trainer and ComfyUI integration workflows, further lowering the barrier to entry.

Industry Impact and Future Prospects

Lightricks' recent update elevates AI video generation to a new level. The breakthrough in 60-second video generation not only enables independent creators, marketing teams, and game developers to quickly generate long-form content but also lays the foundation for real-time interactive applications, such as dynamic AR visuals or player-driven game cutscenes.

Social media feedback shows users are impressed with the speed and quality of LTXV. For instance, one user reported generating a 768x512 video on an RTX 4080 in 45 seconds, and the 60-second video output was astonishing, although some users mentioned that workflow setup still needs improvement.

However, the non-commercial use restrictions of LTXV (limited to academic and research purposes) may pose challenges for some commercial creators. Lightricks stated that a commercial version of LTXV supporting 60-second videos will be launched on its flagship platform, LTX Studio, further expanding application scenarios.

AIbase believes that LTXV's update redefines the possibilities of AI video creation. Its open-source strategy and efficient performance will accelerate industry innovation. In the future, as the model continues to improve, LTXV is expected to play an important role in professional film production and real-time content generation.