Skip to content
8 min read

How to brief AI for images that don't look AI

The Maker track is about the tiny craft of asking — the part that separates your-looking-yours from generic.

There’s a sticky visual aesthetic that has started showing up everywhere — soft-lit, slightly airbrushed, beautifully bland. It’s the default output of the current generation of image models when you give them a vague brief. The Maker track is about learning to avoid it.

The core insight

The entire difference between an AI image that looks like your work and one that looks like stock footage is the brief. A brief of “make this cover look cool” gets you aesthetic wallpaper. A brief of “black-and-white, high-contrast, reminiscent of Nan Goldin’s early 90s portraits, subject centred, slight grain” gets you something usable. This sounds obvious. It’s the one skill that takes the longest to internalise, because the model will happily accept your vague brief and produce something that looks good enough to discourage you from writing a better one.

The daily loop

A Maker quest is 10 minutes against a real visual job:

  1. Pick the asset (cover, thumbnail, icon, illustration, photo edit).
  2. Write a brief before you touch a model: references, palette, mood, the three things it must and must not be.
  3. Generate four; reject three.
  4. Iterate on the survivor with specific edits, not vague “make it better.”
  5. Save the final brief to your personal style doc.

Three examples

Album cover / podcast art. Brief: “Photo-real, single subject, reminiscent of Saul Leiter’s street work, warm tones, heavy grain, a hint of blur at the edges.” Rejects anything that looks like a watercolour.

LinkedIn header. Reference your actual background — your real desk, your real city, a real typeface you use. Use AI to compose, not invent from nothing.

Icon set for a personal project. Brief the weight, the rounding, the stroke style. Do all six at once so they match. Don’t iterate one at a time; the set drifts.

What to watch for

  • Adjective soup. “Clean, modern, bold, vibrant, dynamic” is the default slop. Each word has to do actual work. If you could delete one without changing the output, you had slop.
  • Scale wobble. AI images are often almost right at small sizes and wrong at full resolution. Always view at target size before deciding.
  • Taste drift. Over months your AI outputs will converge on the model’s aesthetic, not yours. Re-read your style doc every week. Add counterexamples.

Where to start

The Maker track opens with a brief-writing drill. You write two briefs for the same job; Voto shows you both outputs; you notice how much one word’s specificity changed. Twelve minutes, real feeling.

Practise this track

Ready to stop reading?

Open the maker track →