LTX-2

Why we included this project

LTX-2 is an audio-video model from Lightricks that generates synchronized sound and picture from a single DiT-based model, so you don't have to bolt a video generator onto a separate audio pipeline. The package includes the official Python inference code and a LoRA trainer, which means you can run it on your own hardware and fine-tune it on your own clips instead of depending on a hosted API. For teams prototyping short-form content, the optional compiled CUDA kernels speed up generation on local GPUs, and the README walks you through pulling the weights from Hugging Face and producing a first output in a short session. It also offers multiple performance modes and production-ready outputs, and the model weights are open access, so it's practical to evaluate without committing to a subscription.