LTX-2 is an open Python inference and LoRA trainer package for an audio-video generative model. It enables text/image-to-video and video-to-video generation, including the production of synchronized audio and video.
Project overview
It provides an open package for DiT-based audio-video generation, including capabilities for text-to-video, video-to-video, and LoRA training.
Project type
Model Runtime · Video · Audio & Speech
Use cases
Video Creation
Deployment
Refer to project documentation
License
License pending
Best for
Developers, creators, and AI engineers who need an open Python inference package and LoRA trainer for audio-video generation.
Key capabilities
Includes pipelines and tools for training and fine-tuning LoRA, full fine-tuning, and IC-LoRA.
Pipelines support automatic prompt enhancement via an enhance_prompt parameter.
Limitations and risks
Inference can be constrained by GPU memory, requiring users to use FP8 quantization or CPU/Disk offloading.
The project requires specialized hardware (GPUs) and manual installation of attention optimizations like FlashAttention. Manual downloading of LTX-2.3 models and the Gemma 3 12b text encoder is required for operation.
Getting started
Setup difficulty is medium, requiring users to clone the repo, download multi-gigabyte models using HF CLI, authenticate with a HF token, and run distilled pipeline commands.
Evidence and sources
GitHub project description: Official Python inference and LoRA trainer package for the LTX-2 audio–video generative model.
README: LTX-2 is the first DiT-based audio-video foundation model that contains all core capabilities of modern video generation in one model: synchronized audio and video, high fidelity,…