About this model
LTX Video 2.0 19B Distilled is the image-to-video member of Lightricks' LTX-2 generation, an open-source, DiT-based audio-video model released in January 2026. Lightricks describes LTX-2 as designed to generate synchronized video and audio in a single pass, removing separate audio post-production steps. The model uses an asymmetric dual-stream architecture in the 19-billion-parameter class, splitting capacity between video and audio, and can produce 4K clips up to roughly ten seconds long with dialogue, ambient sound, and music.
This distilled checkpoint is the speed-optimized counterpart to the full-quality LTX Video 2.0 19B. According to Lightricks' documentation, the distilled path applies a distilled LoRA and needs far fewer inference steps, while the full model runs more steps for finer detail and more nuanced motion. That trade-off makes the distilled variant a practical starting point for rapid iteration and experimentation.
It sits alongside same-generation siblings such as the LTX Video 2.0 19B Distilled text-to-video model and the earlier LTX Video 2.0 Full Quality line. Both full and distilled checkpoints derive from a shared base and can be combined in a multiscale pipeline with spatial and temporal upscalers.
Lightricks has since shipped a later generation, LTX Video 2.3 Fast (March 2026), with a distilled variant designed for low step counts. The toolkit—weights, training code, and ComfyUI workflows—remains openly available.
This About section is AI-generated from public sources (Claude Opus 4.8), with no human editing. It may contain inaccuracies — verify critical details against the sources listed above.
Data sources: Venice API · HuggingFace · Wikipedia — enrichment updated 1d ago