lightricks/ltx-video

LTX-Video is an open-source video generation system built by Lightricks, the company behind the Facetune and LightLeap apps. At its core, it is a machine-learning model that takes a text description or an image and produces a short video clip. You type something like "a cat walking through autumn leaves" and the model renders it as moving footage. The system supports several generation modes. Text-to-video takes a written prompt and creates video from scratch. Image-to-video takes an existing photo and animates it, adding motion that fits the scene. You can also provide multiple keyframes, letting the model fill in the footage between two images you supply. There is a video extension mode that takes an existing clip and continues it forward or backward in time. Model sizes range from a 2-billion-parameter version, which runs on less powerful hardware, up to a 13-billion-parameter version aimed at higher quality results. Both come in distilled variants, which are faster to run, generating a preview in a few seconds on professional-grade hardware. The README describes output up to 4K resolution at up to 50 frames per second. A newer model called LTX-2 adds synchronized audio generation alongside the video, though LTX-2 lives in a separate repository. For people who want to run it locally, installation uses Python and pip, and the model weights are downloaded from Hugging Face. There is also integration with ComfyUI, a popular node-based visual tool that many AI image and video creators already use, which means you can slot LTX-Video into existing workflows without writing code. A set of control models for depth, pose, and edge detection let you constrain how the generated video looks, for example keeping a specific body pose across frames. The project is licensed under a custom Lightricks license visible in the repository. It is aimed at developers, researchers, and technically experienced creators who want to run video generation on their own machines or build it into their own tools.