I have no eye for visual design. I’m terrible at choosing colors that go together or arranging things in pleasing ways. The nicest thing you can say about my PowerPoint slides is that they’re…functional.
Lately I’ve been experimenting with creating visuals with ChatGPT. I didn’t want to create stuff that looked like all the generic AI-generated graphics out there—you know what I mean: Colorful, highly polished, with three-line decorative bursts surrounding everything like little jazz hands.
Yikes. The key to avoiding generic visuals is not to leave the design entirely up to AI.
So what do you do when you’re not a designer yourself? Try these strategies:
1. Provide at least some design direction.
I know I JUST said I wasn’t a designer, but that doesn’t mean I should leave all the design decisions to ChatGPT. Even if you don’t know the terms for things, you can describe them.
“Make it clean and professional” is too vague to be effective. This is how you end up with something that looks generic or just doesn’t work.
A prompt like this is better:
Make this simple and easy to scan. Use a light background, dark text, and one or two muted colors. I want the headline to stand out clearly. Leave plenty of empty space and don’t cram things together. Avoid lots of boxes, generic icons, gradients, or decorative elements. Keep it friendly and approachable, but not cutesy.
If you’re including text with your image, tell ChatGPT where to focus. “This is the main point. These are three supporting points.”
2. Give it something specific to react to.
If (like me) you often don’t even know where to begin, start with something that already exists. You know what you like and don’t like. Find an example of an image that appeals to you, show it to ChatGPT, and explain what you like about it. You don’t need to know the design terms. Just describe what looks good to you.
The goal isn’t to copy someone else’s design. You’re identifying what you respond to so ChatGPT can use those ideas to create something new.
3. Critique the result.
So you created the first draft. How did ChatGPT do with your instructions? Does the image have that generic AI look? Ask yourself:
- Does everything have a box around it?
- Does everything look equally important?
- Are there icons everywhere?
- Is it decorated rather than designed?
- Could this graphic have been created by anyone?
If the image isn’t as distinctive as you’d like, add features that come from you: your colors, a font you like, a different illustration style, or images that make it stand out.
(I’ve mentioned this before, but ChatGPT works best when you change one element at a time. Tell it one thing you want it to change and add, “Don’t alter anything else about the image.”)
4. Subtract before adding.
ChatGPT is a maximalist. It loves to create visually busy images. It often helps to take things away before changing them around. You can ask ChatGPT to remove unneeded icons, borders, labels, footers, or gradients. Try limiting the color palette to one or two main shades. And remove teeny text that’s hard to read.
5. Recognize when you’ve hit the point of diminishing returns.
You’ve got to know when to stop. At some point, another round of prompting won’t make the visual better. AI image editing isn’t terribly precise, so continuing to tweak with additional prompts can lead to frustration and waste your time.
Once the image is close enough, stop asking ChatGPT to perfect it. Use it as-is, or make the last small fixes yourself in Canva, PowerPoint, or another tool.
Designing with ChatGPT isn’t the same as working with a professional designer. But the result may be better than anything you could have created on your own. (In my case: it’s definitely better.) And the idea isn’t to try to fool people into thinking you used a professional to create the image. Make a good faith effort to create something useful that looks nice, and most people won’t care that you used AI to fill in the gaps in your own skills.