Recently, I was walking through some designs with Finn. He had incorporated stunning photography that fit perfectly with the product’s design. I asked him where he had sourced it from, fully expecting to learn about some elite stock photography site. But Finn had generated the images in Midjourney—the most evocative, natural, and usable AI-generated images I’d ever seen.
Getting consistent, high-quality output from AI image generators can feel a lot like luck. You can get much better results by understanding how the AI interprets your words and structuring your prompts strategically.
What this essay explores
- Structure prompts around a specific subject, key details, setting, lighting, and style references.
- Use word order, natural time-of-day language, and seasons to establish focus and mood.
- Use cultural cues and film stock references for more consistent aesthetic control.
- Avoid clichéd LLM prompts, generic descriptors, contradictory styles, and overly complicated descriptions.
- Specify representation deliberately instead of relying on biased defaults and generic personality cues.
- Start with a simple concept, refine from the output, and develop your visual taste.
Prompt: “office chair.”
Key takeaway
A guide to writing prompts to generate AI images that don't look so... um... AI generated.
