AI image generators like Stable Diffusion, Midjourney, or Dall-E can produce amazing results. But as anyone who has tried these tools will tell you: It’s not as simple as pushing a button.
If you want to get good results reliably and regularly, you have to deal with the idiosyncrasies of these tools. As with AI text generators, vague prompts will bring forgettable and random results.
Here are some general tips for creating better prompts:
- Use meaningful language. Describe in detail what you want to see. This includes colors, textures, lighting, and composition.
- Styles and techniques. Think about the style you want the result to have. This can include art movements, materials used, or the type of photo.
- Theme and context. Explain the main subject of the image and, if necessary, add contextual information such as location or time of day.
- Composition and Perspective. A request for a close-up will yield a different result than an aerial view.
- Mood and feelings. Yes, you can also add what mood or feelings you want the image to convey. Is it more humorous or melancholic?
As you can see, ideally you already have an idea of what you want the result to look like. However, you don’t need to have all of the above in mind for every image.
At the same time, the more precise your idea is in your head, the more frustrating the process can be. This is because today’s AI tools stumble, for example, on image ideas for which there are not enough (or any) examples in their training material.
Over time, you will learn what works reliably well and what doesn’t.
A few more tips on how to proceed at this point:
- Depending on the AI, you can use an existing image or photo as a template. Even sketches can be sufficient. This way, you determine the composition of the image or a certain posture.
- Step by step, you will gain a better understanding of the tool’s strengths and weaknesses.
- Every tool is different, and even a new version may have a different focus than you are used to. So experiment as much as you can!
- Look around for good examples and learn the terms that lead to the desired results. Don’t copy everything, though, because some AI pros use far too long prompts, because they’ve never tested what really works and what doesn’t. At Ideogram, for example, you can see what others are generating and with which prompts.
The providers of these tools themselves have realized by now that many users have difficulty putting their image idea into words.
That’s why Dall-E, for example, uses ChatGPT: you tell the chatbot what you need as precisely as possible, and it converts that into an appropriate prompt. Ideogram also does this automatically, always showing you which extended prompt it used.
And Google is experimenting with a user interface that lets you choose alternatives to your chosen terms from dynamically generated pull-down menus. This will help you come up with new ideas.
In this respect: In the not-too-distant future, your prompting skills may not be as important as today, because the tools will actively assist you. However, I believe that even then it will still be a good idea to understand these details in order to work towards specific results.