PixVid lifts latent video diffusion into pixel space
November 11, 2026
PixVid replaces the VAE decoder in pretrained latent video models with a convolutional head to generate video directly in pixel space. This approach enables higher resolution and better controllability by avoiding the constraints of latent space compression.
HOW THIS AFFECTS YOU
●
builderThis could lead to video generation tools with better frame-by-frame control.
●
researcherYou can use this method to bypass the resolution and temporal limits of VAE-based latent video models.
●
designerYou may soon have access to much higher resolution video generation tools.