Methodology

This goal of this site is to provide a represenataion of the differences between the most popular image generation models. The inspiration for this project came when I was getting started with image generation and I was wondering which model performed the best, but found few resources to compare models.

I've come to learn that there is a lot of skill to getting a good image from each model, and each model has it's own ways of interpreting a prompt. Although each model has it's quirks, I hope this website can be a good example of the baseline performance of each model.

Since finetuning each model can result in drastically different results, I've decided to use the default settings on every model. Additionally, I used the first result from each image generation, regardless of how many images the model generated. I did this to avoid my bias in selecting the "best" image from each generation.

Model Notes:

DALL·E2

  • Plugged each prompt direclty into  labs.openai.com
  • Used the first image (top left) each time

Stable Diffusion

Midjourney

  • Used the Midjourney Discord to generate images
  • Did not used v5 since this is not the default option as of now
  • The only service where I was not able to generate the images for free with trial credits

All of these services were fairly difficult to use since every service said it was facing high demand and took multiple tries to generate the images. Stable Diffusion offers significantly more customization than both models, but I did not do any of this.

If you have any questions/ comments about the methodology used, pleae let me know!