Skip to content

I used the online CogView4 model by Zhipu AI to generate an image. The result matched the expectation, but the text in the image was all in English instead of the specified Chinese.

I haven't tested the open-source version, but the online version should theoretically be more powerful.

It seems the prompt was too complex, and the model couldn't understand or follow it. Or is English still prioritized internally?

Zhipu AI Entry https://bigmodel.cn/trialcenter/modeltrial

Here is the prompt

Please draw a picture:
### Overall Layout
- Simple cartoon style
- The image is divided into two parts, "Before OpenAI" on the left and "After OpenAI" on the right, connected by an arrow (→).
- Each part contains two scenes (top: coding, bottom: fixing bugs)

### Left: Before OpenAI
1. **Top: Developer Coding**
   - Background: A simple desk with an old-fashioned computer monitor.
   - Character: A cartoon developer (round head), sitting in front of the computer, looking focused and a bit confused.
   - Text: Write "Developer Coding - 2 hours" in a bubble above the developer's head or at the top of the frame.

2. **Bottom: Developer Debugging**
   - Background: Same desk and computer, but the developer looks tired and frustrated, holding his head in his hands, staring at the computer screen.
   - Character: Same cartoon developer, looking pained.
   - Text: Write "Developer Debugging - 6 hours" in a bubble above the developer's head or at the top of the frame.

### Right: After OpenAI
1. **Top: ChatGPT Generates Code**
   - Background: Same desk and computer, but there may be an icon indicating ChatGPT next to the computer screen.
   - Character: The developer is sitting in front of the computer, looking relaxed or surprised, indicating that the code has been generated by ChatGPT.
   - Text: Write "ChatGPT Generates Code - 5 minutes" in a bubble above the developer's head or at the top of the frame.

2. **Bottom: Developer Debugging**
   - Background: Desk and computer, the developer looks even more tired and desperate, holding his head in his hands.
   - Character: Same cartoon developer, looking more painful.
   - Text: Write "Developer Debugging - 24 hours" in a bubble above the developer's head or at the top of the frame.

Actual generated images

Testing with a simple prompt, the effect is good

It seems that for complex, multi-scene images, at least text embedding support is not yet in place. However, for simple scenes, especially beaches and advertisements, the results are very good.