I don't need paragraphs in my prompts. I need a core. Usually that's 3–7 words and parameters that keep the frame in check. That way I can see exactly what I'm changing — and I stop confusing causes with effects.
My "why"
- Clear intent — the prompt core is what stays. Everything else is for experimenting.
- Repeatability — one change at a time. I know what worked.
- Hygiene — short prompts are readable the next day. I don't dig through essays.
My rule: first I see the image, then I write a short core + parameters. I iterate one change at a time.
The structure I use
- Subject (what it is): e.g. brutalist tower, editorial portrait, watch product shot
- Technique / mood: 1–2 words (e.g. soft light, overcast, kodak portra)
- Parameters: --ar (ratio), --stylize (intensity), --chaos (variety)
Examples from my workflow
brutalist tower, overcast, cinematic — --ar 16:9 --stylize 150editorial portrait, soft window light, kodak portra — --ar 4:5 --stylize 200watch product shot, glossy reflections — --ar 1:1 --stylize 120
Parameters I always have ready
- --ar — I match it to the medium (Instagram: 4:5/1:1, web: 16:9, stories: 9:16).
- --stylize — 120–300 is my "safe zone". I go higher when I want more character.
- --chaos — 0–20 for stability, 20–40 when exploring, 60+ when I want something strange.
The most common mistakes I've cut out
- Wall of text — the model gets distracted. So do I.
- Mixing five aesthetics at once — "portra + ektachrome + ilford" in one prompt isn't my approach.
- Missing aspect ratio — just --ar alone can fix the whole composition.