Skip to content

Snap Video

A video-first model for text-to-video synthesis

Visit official site

What Snap Video is

The model is described as a video-first approach that extends the EDM framework to address motion fidelity, visual quality and scalability in video generation. The material describes a scalable transformer architecture that treats space and time as one highly compressed 1D latent vector.

What you can do with it

  • Extend the EDM framework for video generation
  • Use a scalable transformer architecture
  • Model space and time jointly
  • Synthesise temporally coherent video with complex motion

Model notes

  • Redundancy between frames is used as input information
  • Space and time are compressed into a 1D latent vector
  • The architecture is described as efficient to train at a scale of billions of parameters
  • The source calls it a scalable spatial-temporal transformer for text-to-video synthesis

Example uses

  • Generate a video from an advertising slogan
  • Generate a short video from a prompt about a cat chasing a butterfly
  • Generate a New Year greeting video from a written prompt

Pros & cons

✓ What we like

  • Joint modelling of space and time in one compressed 1D latent vector
  • Architecture described as efficient to train at a scale of billions of parameters

! What to watch out for

  • The approach is built on extending the EDM framework
  • Video output is driven by a compressed 1D latent representation

FAQ

What framework does it extend?

The material states it extends the EDM framework for video generation.

How are space and time handled?

The material describes a highly compressed 1D latent vector for joint spatial and temporal modelling.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review