Skip to content

SF-V

Single-step video generation from a diffusion model

Visit official site

What SF-V is

The model is a diffusion-based video generator tuned with adversarial training, able to synthesize high-quality video in a single forward pass.

What you can do with it

  • Generate video through one forward pass
  • Fine-tune a pretrained video diffusion model with adversarial training
  • Initialize the generator and discriminator from image-to-video diffusion weights

Pros & cons

✓ What we like

  • One-step generation lowers denoising cost
  • Video comparisons and ablation analysis are provided

! What to watch out for

  • Code and demos are listed as coming
  • Training freezes the UNet encoder part

FAQ

How is it trained?

By adversarial training on a pretrained video diffusion model.

What gain is reported?

About 23 times faster with better quality.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review