LTX Video

LTX Video (LTXV) is an open-source diffusion transformer (DiT) video generation model developed by Lightricks, the studio behind Facetune and LTX Studio. It generates text-to-video, image-to-video, and video-to-video sequences, with the LTX-2 generation adding synchronized audio (dialogue, motion sound) produced in the same pass as the visuals rather than dubbed on afterward. The model targets creators who want local, license-clean control over generative video rather than being locked into a closed API. Weights ship under the permissive OpenRail-M license, which allows commercial use, and the repo has become one of the most-starred open video models on GitHub with tight integrations into ComfyUI, Diffusers, Fal, and Replicate. Multiple variants are published: a 13B full model for maximum quality, a 13B distilled build for faster inference at similar fidelity, a 2B distilled model for rapid iteration on consumer GPUs, and FP8-quantized checkpoints that cut VRAM further. Feature-wise, LTXV is notable for multi-keyframe conditioning (you can pin the first, middle and last frame and let the model tween), forward and backward video extension, and native 4K generation at up to 50 FPS. Typical workflows include storyboarding a short film in LTX Studio, running local ComfyUI pipelines for controlled shots, using it as a Sora/Runway alternative inside custom pipelines, and building bespoke video tooling on top of Fal or Replicate endpoints. Because it is a research-grade open model, output quality still trails the best closed systems on complex prompts, and running the 13B model well requires a serious GPU (roughly 24GB VRAM and up for comfortable use).