Skip to content

CogView3

A cascaded diffusion system for text-to-image generation

Visit official site

What CogView3 is

The system is described as a text-to-image generator using cascaded diffusion, where high-resolution output is split into stages with a relay super-resolution step.

What you can do with it

  • Generate 512x512 images from text
  • Apply 2x super-resolution to the result
  • Extend generation up to the 512 to 2048 range

Model notes

  • Gaussian noise is added to low-resolution results
  • Zero-SNR diffusion noise scheduling is listed
  • Joint text-image attention and a 16-dimension VAE are named
  • The material states faster generation and higher image quality than SDXL

Pros & cons

✓ What we like

  • A 2x super-resolution stage is listed
  • Joint text-image attention and a VAE are named

! What to watch out for

  • Generation is split across multiple stages
  • Diffusion starts from low-resolution results with added noise

FAQ

What framework does it use?

The material describes cascaded diffusion with a relay diffusion framework.

What resolutions are listed?

512x512 generation, 2x super-resolution and a 512 to 2048 range are listed.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review