A few months ago, I ran an experiment to create Telex cover images using OpenAI and Google Nano Banana models. The goal was to generate images with a specific style using the same prompt across both models.
Telex is an experimental AI tool for creating WordPress blocks through an interactive AI coding assistant in a visual, live-preview environment directly in the browser.
The idea behind the experiment was to reproduce the dithering aesthetic of the Telex website for user-created projects. These cover images would be used as sharing images for blogs and social platforms, and potentially in the future dynamic project showcase on the Telex homepage.
These are some of the images produced. All of them were created using the same prompt:
Example project: “Confetti button block”
Google vertex imagen-4.0-generate-001
Open AI gpt-image-1










Example project: “Contact form block”
Google vertex imagen-4.0-generate-001
Open AI gpt-image-1










Example project: “Hosting pricing block”
Google vertex imagen-4.0-generate-001
Open AI gpt-image-1










Some conclusions after comparing both models
| Model | Google vertex imagen-4.0-generate-001 | Open AI gpt-image-1 |
|---|---|---|
| Speed | ~ 9 seconds per image | ~30 seconds per image |
| Color consistency | Consistent: always pure black and white | Not consistent: Sometimes the white is yellowish |
| Dithering | Not real dithering effect but more consistent type of dots across images | Not real dithering effect and inconsistent type of dots across images |
| Subject | Seems more abstract and not so literal | Seems more figurative and literal |
Leave a Reply