Skip to content

CogView

Generate images from plain language with a Transformer

Visit official site

What CogView is

The model is described as a pretrained Transformer for general-domain text-to-image generation with 41 billion parameters.

What you can do with it

  • Generate matching images from plain language
  • Enter descriptions in Chinese or English
  • Improve quality with super-resolution

Application scenarios

  • Generate an image of a chubby cat on a table
  • Generate a pink rose blooming in sunlight
  • Generate white clouds drifting in a blue sky

Pros & cons

✓ What we like

  • Super-resolution improves generated image quality
  • Generated samples can be filtered afterwards

! What to watch out for

  • Training uses PB-relax and Sandwich-LN for stability
  • Fine-tuning targets a specific domain after pretraining

FAQ

How many parameters?

The material states 41 billion parameters.

Which input languages?

The material lists Chinese and English.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review