When you work with LLMs like ChatGPT, long, complex prompts with lots of context and detail are often more effective than simple prompts. But when you work with AI image generation tools, those long prompts often don’t work as well. Most image generation tools give better results with simpler prompts rather than long paragraphs. I use a standard prompt structure for AI image generation. The structure isn’t strict; I shift elements around depending on what’s most important. In my experience, this prompt structure works well across multiple AI image tools.
Prompt structure
This is my basic structure for AI image prompts:
Image styleSubjectActionPhysical characteristicsClothesSetting/environmentAdditional details
The order is somewhat fluid, especially for the action, physical characteristics of the subject, clothes, and setting/environment. You also don’t need to include every element in this structure for every prompt.
Example prompt
Let’s break down an example prompt to see this structure in action.
Editorial photo, Asian woman working on a laptop in a sunny home office, focused expression, blue coffee mug on the desk
Prompt with phrases, not sentences
Image generation prompts typically work best with short phrases separated by commas rather than complete sentences. When you prompt with full sentences or paragraphs, the image generation tool has to disregard all of your filler words. It’s more likely to get confused about what’s important in your image when it has to find the key phrases within longer sentences.
ChatGPT is one exception to this guideline. ChatGPT’s image generation works differently than tools like Midjourney, so it’s much more forgiving of longer prompts and sentences.
Important elements first
In your prompt structure, put the most important elements first and the less important elements at the end of the prompt. Image generation tools usually prioritize the elements listed at the beginning of your prompt. Noting the image style first, as in the prompt structure I use, sets the tone for how the tool interprets the rest of the prompt details.
You can play around with the order of the elements though, especially in the middle of the prompt. For this example prompt, I was primarily interested in getting the details of the woman’s appearance right. Therefore, I listed the details about her age, hair, and clothes before I listed the action.
Editorial photo, Pakistani woman, mid 50s, slightly graying hair pulled back in a low knot, wearing a red and yellow sari, gold chandelier earrings, talking and gesturing with her hands, modern office conference room
In this example, the illustration style was important, so I front-loaded my prompt with two phrases about the style (“single line vector graphic” and “continuous black line”). I also included the additional reinforcement phrase of “black and white” at the end.
Single line vector graphic, continuous black line, people sitting at a table for a meeting, laptops, work, black and white
Experiment with your prompt structure
While I use this structure as a starting point for most of my prompts, it’s worth experimenting with these elements. Try changing the order or expanding certain sections of the prompt.
While I’ve had better success using a series of short phrases, you may have better luck with longer prompts, especially for certain images or with different tools. Use this structure as a jumping off point, not a rule to strictly follow.
Resources
Rory Flynn is one of the people I follow to learn how to use Midjourney better. Much of the structure I use in my image prompts is based on his examples and tips. He offers a number of free guides for how to learn Midjourney on his website. If you want a much deeper dive into image prompts, start with his resources.
Google’s Say What You See is an interactive guide to improving your skills at describing images, an important skill for prompting.
For more examples and tips for prompts, check out my blog posts on AI image generation.
