Skip to content

FouriScale

Generate high-resolution images without extra training

Visit official site

What FouriScale is

The method is described as a frequency-domain view on high-resolution image synthesis without extra training.

What you can do with it

  • Generate high-resolution images
  • Handle varied aspect ratios
  • Reduce repeated patterns

Model notes

  • Original convolution layers are replaced
  • Dilation and low-pass operations are used
  • Padding then cropping is applied

Pros & cons

✓ What we like

  • No training is required
  • Structure and fidelity are balanced

! What to watch out for

  • It builds on pretrained diffusion models
  • Outputs depend on the input prompt

FAQ

Is training needed?

No, the method is described as training-free.

What is changed?

Original convolution layers are replaced.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review