LTX-Video
Open-source DiT video model generating HD clips in seconds, from 2B to 13B sizes.
About
LTX-Video (LTXV) is Lightricks' open-source diffusion-transformer video generation model, notable as the first DiT video model capable of near-real-time generation. The 13B distilled variant generates 1216x704 video in under 10 seconds on a consumer RTX 4090, roughly 30-40x faster than comparable models, while the 2B distilled version targets lower-VRAM machines.
Released fully open source on GitHub and Hugging Face, it supports text-to-video, image-to-video, keyframe editing, camera motion control, multi-shot sequencing, and multi-condition generation. Lightricks also trains on licensed data via Getty and Shutterstock partnerships to reduce enterprise copyright risk. Limitation: quality still trails closed flagship video models on complex motion, and the 13B dev model needs serious GPU memory without quantization.
What it can do
- Text-to-video and image-to-video generation
- Near-real-time inference with distilled variants
- Keyframe editing and camera motion control
- Multi-condition generation across frames
Setup
Clone github.com/Lightricks/LTX-Video or pull weights from Hugging Face; ComfyUI nodes available.
Prerequisites
NVIDIA GPU with CUDA (RTX 4090 recommended for 13B); Python 3.10+ and PyTorch.
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 65; Trending score 85/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.