mountainplatform300/ltx-desktop

LTX Desktop is a desktop application for generating videos with AI, using the LTX family of video models. This particular repository is an unofficial fork of Lightricks's original LTX Desktop project, kept up to date with credit to the original authors while adding an integrated workflow for training custom LoRA models, which are small add-on files that teach a video model a specific style or subject. The app lets a user turn text, an image, or audio into a video, and also includes a video editor, editing projects, and a Retake feature for regenerating parts of a video. It supports creating and training LoRA datasets locally or through a cloud GPU service, and includes a library for managing trained LoRAs, example prompts, and previews. How it runs depends on your computer. On Windows or Linux with an NVIDIA graphics card that has at least 16 gigabytes of video memory, the app can generate videos locally. Any other setup, including all Mac computers, requires an API key from LTX's cloud service instead, since the video generation then happens on their servers rather than on your machine. Local generation needs a fairly capable machine: 16 or more gigabytes of RAM and over 150 gigabytes of free disk space for the model files. Installation is done through downloadable installers for Windows, macOS, and Linux, though the current beta release is not yet digitally signed, so the operating system will show security warnings during install. Advanced users can instead build the project from source using Node.js, pnpm, Python, and a tool called uv. Under the hood the app is built from three parts: a React and TypeScript interface, an Electron layer that handles the operating system integration, and a Python backend built with FastAPI that manages the actual video generation and talks to external APIs only when needed. The project is in beta and the interface is under active redesign, so the maintainers are currently limiting large interface contributions.