[Papers] LTX-Video: Realtime Video Latent Diffusion - research promises to deliver even faster video generation and real-time capabilities.

Researchers have introduced LTX-Video, a transformer-based latent diffusion model for video generation that integrates the Video-VAE and denoising transformer to improve efficiency and quality. The model achieves a high compression ratio of 1:192 and supports text-to-video and image-to-video generation, producing 5 seconds of 24 fps video at 768x512 resolution in 2 seconds on an Nvidia H100 GPU. Its source code and pre-trained models are publicly available. Open article →