LTX-Video: open model generates 5s of video in 2s on a single H100
Crazy-Repeat-2006 · reddit · 2026-11-10
LTX-Video is a transformer-based latent diffusion video model that jointly optimizes the Video-VAE and the denoising transformer instead of treating them independently.
- The VAE achieves a 1:192 compression ratio (32×32×8 pixels per token) by moving patchifying from the transformer input to the VAE input.
- This highly compressed latent space lets the transformer run full spatiotemporal self-attention efficiently for temporally consistent high-res video; the VAE decoder handles latent-to-pixel conversion plus a final denoising step in pixel space, preserving fine details without a separate upsampler.
- Text-to-video and image-to-video are trained simultaneously. It generates 5 seconds of 768×512, 24fps video in just 2 seconds on a single H100 — faster than real time and outperforming similar-scale models.
- Source code and pretrained weights are publicly available.
Related event: LTX-Video Generates 5-Second Video in 2 Seconds on a Single H100(2 posts)→
More from Multimodal
- ComfyUI long-video node breaks Viggle Remake's 345-frame limit with overlapping windows — Hoje-Na-IA · 2026-11-10
- Research: stitching video from multiple moving cameras with a visually pleasing alignment net — kwangmoo_yi · 2026-11-10
- Claude Plus Blender MCP Codes a 931-Part, 24,260-Rivet Jet Model — gaganghotra_ · 2026-11-10
- Claude Fable 5.2 Writes and Records a Beethoven-Style Piano Sonata in LilyPond — imjustnewatai · 2026-11-10
- Fable 5.2 impresses with motion graphics, cooking up a Convex explainer from one prompt — Rasmic · 2026-11-10
- Fable 5.2 recreates Avengers' Battle of New York in 3D, scene-for-scene with the movie — every · 2026-11-10