LTX-2.5

LTX-2.5 is an open model from Lightricks that generates video together with sound: picture and audio come from a single model. It is built on a 22-billion-parameter transformer with a built-in text encoder; the repository includes distilled and full versions, spatial and temporal upscalers, fp8, int8 and nvfp4 quantizations, and LoRA. Modes include text-to-video, image-to-video and other pipelines listed in the author's repository. The weights are published on Hugging Face; to download them you need to accept the terms on the model page. The license is the LTX-2.x Community License Agreement: per the license text, companies with annual revenue of 10 million US dollars or more need a paid license from the author for commercial use; the full license text is in the author's GitHub repository. The author does not state a VRAM figure; the documentation describes quantization and memory offloading modes for smaller cards. A demo without installation is the playground on the author's site.