Prompt Engineering for Generative AI

Chapter 8. Standard Practices for Image Generation with Midjourney

In this chapter, you’ll use standardized techniques to maximize the output and formats from diffusion models. You’ll start by tailoring the prompts to explore all of the common practices used for image generation. All images are generated by Midjourney v5, unless otherwise noted. The techniques discussed were devised to be transferrable to any future or alternative model.

Format Modifiers

The most basic practice in image generation is to specify the format of the image. AI image models are capable of deploying a wide variety of formats, from stock photo, to oil paintings to ancient Egpytian hieroglyphics. The image often looks completely different depending on the format, including the style of the objects or people generated in the image. Many of the images in the training data are stock photos, and this is also one of the most commercially important image categories for image generation.

Input:

a stock photo of a business meeting

Figure 8-1 shows the output.

Figure 8-1. Stock photo of a business meeting

The ability to generate infinite royalty-free stock photos for free with open source models like Stable Diffusion, or for a very low cost with DALL-E or Midjourney, is itself a game changer. Each of these images is unique (though may contain similarities to existing images), and therefore they look more premium ...