Many folks just starting with AI are surprised when they don’t get the output they expect. Chances are, the prompt given to the AI is not specific enough for it to “see” what you have in mind. This post will provide some general tips on constructing an image generation prompt, as well as a link to a free tool for practicing your prompts in a unique way.
Building a Prompt
A good prompt for an AI that generates images contains three basic parts:
Subject matter: What do you see? Pair nouns and adjectives to describe the main subject. It xan include information about the setting. Examples: leafy green plant, shiny iPhone, angry orange cat, on a beach, in a tree, etc.
Action: What about it? Add in a verb to tell what the subject is doing. Examples: sitting, running, studying, jumping, floating, laughing etc.
Modifiers: What does it look like? You can put in a mood, art medium, or art movement. Examples: photograph, pop art, gloomy, manga, pen and ink, impressionism, etc. Here is a huge list of modifiers to get you started. Not sure what some of them mean? Here is a good source of examples of art movements and styles.
Practice Prompting
Now that you have the basic idea, it’s time to practice! You could start throwing prompts out there and hope for the best, but that could be frustrating. Google’s Say What You See tool shows you an image and asks you to describe it. When you enter your description, it generates a new image based on your description (prompt) and compares the images side by side. That way, you can see which details were needed to make your image look more like the original. Sometimes the tool misbehaves by not doing what you asked, but for the most part, it’s a helpful way to see the difference between what you’re describing and what the AI “hears.”
