Lightricks' fast open-weights video model that generates 5-second clips in near real-time on consumer GPUs.
LTX Video (LTXV) from Lightricks is the first video model fast enough to matter for actual production work — generating 1216x704 clips quicker than they play back, on a single consumer GPU. Open-weights, ComfyUI-native, and with a distilled variant that cuts inference to seconds. While Sora and Veo chase cinematic polish behind a paywall, LTXV quietly became the default for iteration-heavy workflows where you need 30 rough drafts before one final render.
Who it's for: Motion designers, YouTubers, and indie creators who want unlimited iteration without per-second render pricing. Also the right answer for teams that need video generation behind their own firewall.
Generates faster than the video plays — a 5-second clip in under 4 seconds on an RTX 4090 via the distilled model.
Apache-licensed weights on HuggingFace. No API key, no rate limit, no content filter beyond what you set.
First-class ComfyUI support with control nodes for depth, pose, and camera motion — slot it into existing pipelines.
The 13B model for quality finals, the 2B distilled model for instant drafts. Match the model to the stage of work.
If you run a video workflow in 2026, LTX Video is your drafting engine. Use it to explore 30 concepts for the cost of one Sora render, then send the winner to Veo 3 or Kling for final polish. The open weights alone make it a permanent part of the stack.