IndexMapAnti-DoomscrollAbout
Model FamilyVideoCross-Modal

LTX-Video & LTX-2

Also known as: LTX-Video · LTX-2 · LTX-2.5 · LTXV · Lightricks LTX

Plain English

Lightricks' family of high-speed open-weight video and audio-visual foundation models, engineered to generate video clips—and in LTX-2 and LTX-2.5, synchronized speech and sound effects—locally on consumer and workstation GPUs.

Technical Overview

LTX-Video couples a high-compression 1:192 spatiotemporal Video-VAE (32x32x8 patch downsampling) with a flow-matching DiT while folding the final denoising step directly into the VAE decoder. LTX-2 (19B) and LTX-2.5 (22B, 4K HDR) extend this design to an asymmetric dual-stream audio-visual DiT distributed globally under a community license that is free for commercial use under $10M in annual gross revenue.

Timeline (5)

RSS
  1. 2024

  2. Paper

    Lightricks published LTX-Video: Realtime Video Latent Diffusion, pairing a 2B-parameter DiT with a 1:192 spatiotemporal Video-VAE whose decoder performs the final denoising pass—allowing 768x512 clips to render faster than their playback duration on an H100 GPU.

  3. 2025

  4. 2026

  5. Release

    Lightricks released LTX-2.5 (22B parameters) with single-pass native audio, multi-shot consistency, and 4K HDR support in ComfyUI and Diffusers under the LTX-2.x Community License—allowing global self-hosting and free commercial use for creators and companies under $10 million in annual gross revenue.

See also (9)

Open full graph
4 more connections