How to brief an AI tool the way you would brief a person
Vague prompts do not produce vague results. They produce confident wrong ones.
If you handed a freelancer a brief that said “make it modern and clean”, you would expect to be disappointed, and you would know whose fault it was. The same brief given to a generative tool produces the same outcome, except it arrives in four seconds and looks finished, which makes the problem harder to see.
A usable brief answers four things, and it does not matter whether the recipient is a person or a model.
What it is for
Where the output will be used, at what size, in what surroundings. “A header image for a newsletter, read on a phone, sitting above a headline in dark text.” That single sentence rules out half of what the tool would otherwise produce.
What must be true
The non-negotiables. Product visible and unmodified. Space at the top for type. No text in the image. Horizontal. These are constraints, not preferences, and they are the ones worth stating first and repeating.
What it should feel like — in nouns
Adjectives like premium, clean and bold mean almost nothing on their own, because they mean something different to everyone. Nouns work far better: the material, the light source, the lens, the era, the place. “Overcast daylight, matte paper, shallow depth of field” is a direction. “Premium and modern” is a mood board nobody has seen.
What to avoid
Say what you do not want, especially the thing the tool defaults to. Most generators have a house style — a certain glossiness, a certain symmetry — and naming it is the fastest way out of it.
When a brief works, save it as a template with the specifics swapped for blanks. Two or three good templates beat a folder of two hundred one-off prompts you will never find again.
Then iterate on one variable
When the result is close but wrong, change one thing and run it again. Rewriting the whole brief each time tells you nothing about which word was doing the work. This is slower for one image and much faster by the fifth.