In a move supporting open development in creative AI, Lightricks has released the full LTX-2 model as open source. The multimodal model, capable of generating synchronized audio and video, is now publicly available with complete weights and a suite of development tools.

This release includes the full model weights, a distilled version for efficiency, and a comprehensive developer toolkit, positioning LTX-2 as a rare example of a production-capable audio-video generator available for public use and modification.

The company states that high-fidelity multimodal models with open weights are uncommon and often not suited for real creative workflows. LTX-2 is presented as breaking this pattern by combining high-quality output with practical efficiency. The model is optimized to run on modern consumer NVIDIA GPUs, aiming to make advanced audio-video generation more accessible without specialized hardware.

What’s Included in the Release

The release is structured as a full development stack:

· Model Weights: The complete weights for both the main LTX-2 model and a performance-optimized distilled version.

· LoRA Adapters: A set of adapters that demonstrate style and task customization, serving as templates for user creation.

· Trainer Framework: A modular system for fine-tuning and extending the model, including data pipelines and multi-GPU support.

Capabilities and Philosophy

According to Lightricks,LTX-2 supports generating up to 20 seconds of synchronized audio and video at 4K resolution with accurate lip sync and motion coherence. It offers creators fine-grained control through multiple input methods, including text, images, and keyframes.

The company framed the open-source decision as a philosophical stance against the trend of locking advanced AI behind closed APIs. Lightricks believes the next significant advances in creative AI will come from community-driven development, allowing researchers and developers to freely experiment, fine-tune, and build new applications on top of the LTX-2 foundation.

Getting Started

All resources are available on GitHub and Hugging Face. Lightricks also provides a integration for ComfyUI and has established a Discord community for developers to collaborate and share extensions.

By releasing LTX-2 openly, Lightricks aims to transform the model from a proprietary tool into a shared foundation for innovation in multimodal AI, inviting the global developer community to explore its potential.