Skip to content

CogView4-6B

Generate high-resolution images from Chinese or English text

Visit official site

What CogView4-6B is

The model is described as a text-to-image generation model developed by the Knowledge Engineering Group at Tsinghua University, built on deep learning to turn text descriptions into high-quality images. The feature list names high-resolution output, multilingual input and efficient inference, and cites DPG-Bench and GenEval among image generation benchmarks.

What you can do with it

  • Generate images from 512px up to 2048px
  • Enter prompts in Chinese or in English
  • Apply model CPU offload and a 4bit text encoder to lower memory use
  • Run with BF16 or FP32 precision and consult the listed model metrics

Example uses

  • Render a red sports car by the sea from a description
  • Draw a traditional festival scene from Chinese text
  • Draw a science fiction scene from an English description

Getting started

  • Install the Diffusers library from source so the model runs
  • Load the pretrained model with CogView4Pipeline.from_pretrained
  • Configure options such as enable_model_cpu_offload and vae.enable_slicing
  • Supply a detailed text prompt describing the scene, colours and objects

Pros & cons

✓ What we like

  • Detailed model metrics and performance data are provided
  • BF16 and FP32 precision are supported

! What to watch out for

  • Running the model requires installing the Diffusers library from source
  • Image output depends on the text prompt supplied

FAQ

Which languages can the prompts use?

The feature list names Chinese and English text input.

What resolutions are listed?

The material states high-resolution generation from 512px to 2048px.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review