LTX-Video: Realtime Video Latent Diffusion

TL;DR

LTX-Video is a transformer-based latent diffusion model that efficiently generates high-resolution, temporally consistent videos in real-time by integrating a high-compression Video-VAE with a denoising transformer.

Contribution

The paper introduces a novel holistic approach that combines Video-VAE and denoising transformer into a unified model, enabling faster and higher-quality video generation.

Findings

Achieves 5 seconds of 24 fps video at 768x512 resolution in 2 seconds.

Supports both text-to-video and image-to-video generation.

Outperforms existing models of similar scale in speed and quality.

Abstract

We introduce LTX-Video, a transformer-based latent diffusion model that adopts a holistic approach to video generation by seamlessly integrating the responsibilities of the Video-VAE and the denoising transformer. Unlike existing methods, which treat these components as independent, LTX-Video aims to optimize their interaction for improved efficiency and quality. At its core is a carefully designed Video-VAE that achieves a high compression ratio of 1:192, with spatiotemporal downscaling of 32 x 32 x 8 pixels per token, enabled by relocating the patchifying operation from the transformer's input to the VAE's input. Operating in this highly compressed latent space enables the transformer to efficiently perform full spatiotemporal self-attention, which is essential for generating high-resolution videos with temporal consistency. However, the high compression inherently limits the representation of fine details. To address this, our VAE decoder is tasked with both latent-to-pixel conversion and the final denoising step, producing the clean result directly in pixel space. This approach preserves the ability to generate fine details without incurring the runtime cost of a separate upsampling module. Our model supports diverse use cases, including text-to-video and image-to-video generation, with both capabilities trained simultaneously. It achieves faster-than-real-time generation, producing 5 seconds of 24 fps video at 768x512 resolution in just 2 seconds on an Nvidia H100 GPU, outperforming all existing models of similar scale. The source code and pre-trained models are publicly available, setting a new benchmark for accessible and scalable video generation.