Vague in, generic out
Ask for "a nice image of a coffee shop" and you will get a coffee shop. It will be warm, wooden, a little bit Scandinavian, shot from about chest height, with a plant. So will the next one. So will the one your competitor generated last Tuesday.
That is not the tool being unimaginative. It is the tool doing exactly what you asked, which was to produce the average of every coffee shop picture it has ever encountered. If you do not specify, something else specifies for you, and that something else is the average.
The fix is not longer prompts. It is a brief. The same thing you would give a photographer or an illustrator, written down in the same order every time so that you notice the gaps.
The six slots
Think of a brief as six slots. You do not have to fill all six, but you should know which ones you deliberately left empty.
Subject. What is actually in the picture, and what is it doing. Not "a business owner" but "a woman in her fifties in an apron, turning a sign in a shop window". Specificity here does more work than everything else combined.
Style. How it is rendered. Photograph, flat vector illustration, watercolour, pencil sketch, 3D render, collage. If you say nothing, you get a glossy in between thing that belongs to no medium at all. Be careful about naming living artists here, for reasons the rights lesson comes back to. Describing the qualities you want is both safer and more controllable than borrowing a name.
Composition. Where things sit and how close you are. Close up, mid shot, wide. Shot from above, at eye level, from below. Subject centred, or off to one side. Portrait or landscape shape. This is also where you ask for empty space, while accepting from lesson one that you may not get it.
Lighting. Soft and overcast, hard midday sun with sharp shadows, warm late afternoon, single lamp in a dark room, bright and even. Lighting is the single most underused word in most people's prompts and it changes the mood of a picture more than the colour palette does.
Mood. Calm, busy, austere, playful, expensive, homely. One or two words. This is the slot that stops a technically correct image from feeling wrong.
Exclusions. What must not appear. No text. No logos. No people's faces. No modern phones. Nothing on the left third. This slot is the one people forget, and it is the one that removes the most annoyance.
If a brief feels too long, read it back and ask which sentence would change the picture if you deleted it. Anything that would not change the picture is decoration, and you can cut it.
โ Weak prompt
Prompt
A nice photo of a bakery for my website.
Output
A warm, wooden bakery interior, shot at chest height, shelves of generic bread, a small plant, soft golden light. Pleasant. Completely anonymous.
Every slot was left empty, so the tool filled all six with the average. The result is not bad, which is the problem. It is unusable because it says nothing about this bakery in particular.
โ Good prompt
Prompt
Photograph. A baker in her sixties in a flour dusted navy apron, sliding a tray of dark rye loaves into an old steel oven. Mid shot from slightly below, subject on the right, the left third of the frame kept simple and uncluttered. Early morning light through one high window, hard shadows, warm tones. Mood: unglamorous, practised, quietly proud. No text anywhere in the image, no visible brand names, no other people.
Output
A specific person in a specific moment, with room on the left where a headline can sit, and a mood that reads as craft rather than lifestyle.
Six slots, all filled deliberately. It is longer, but every clause is doing a job, and the exclusions at the end save an entire round of regenerating.
Checkpoint
A usable brief covers subject, style, composition, lighting, mood and exclusions, so the tool is not left to fill the gaps with the average.
A template you can reuse
Keep this in a note. Fill the square brackets, delete any line you genuinely do not care about, and paste it in.
STYLE: [photograph / flat vector illustration / watercolour / pencil sketch / 3D render]
SUBJECT: [who or what, and what they are doing, in one specific sentence]
COMPOSITION: [close up / mid shot / wide], [from above / eye level / from below],
subject positioned [centre / left / right], [portrait / landscape / square] shape.
Keep [which area] simple and uncluttered.
LIGHTING: [soft overcast / hard midday sun / warm late afternoon / single lamp in a dark room]
COLOUR: [dominant tones, in words. Exact brand values get set afterwards in an editor.]
MOOD: [one or two words]
DO NOT INCLUDE: text or lettering of any kind, brand names or logos,
[anything else that would ruin it].
Change one thing at a time
The most common mistake after a vague brief is an impatient rewrite. You get something almost right, so you change the lighting, the angle, the mood and the colour all at once, and the next image is worse in a way you cannot explain.
Treat it like seasoning. One adjustment, look, one adjustment, look. It feels slower and it is dramatically faster, because you learn which words are actually moving the picture. After a dozen goes you will have a personal sense of which slot to reach for, and that sense is the real skill in this lesson.
It also helps to keep the version you liked. Save it before you start altering the brief. There is a particular flavour of misery in chasing an improvement for twenty minutes and being unable to get back to the good one.
When the brief is right and the picture still is not
Two things worth trying before you conclude the tool cannot do it.
First, describe the picture rather than the concept. "Trust" is not a thing that can be drawn. Two people looking at the same screen and one of them pointing is a thing that can be drawn. Abstract nouns are where briefs go to die, so translate them into something visible before you type them.
Second, hand over a reference image if the tool accepts one. A picture of the look you are chasing does more than a paragraph describing it, and it removes the guesswork about what you meant by "clean" or "moody" or "editorial". Use a reference for style and framing. Do not use a competitor's finished artwork and ask for something close to it, which is a rights problem rather than a craft problem.
Watch for briefs that quietly ask for something factual. "A diagram of how our returns work", "the interior of the Royal Albert Hall", "our product from the back". The tool will produce something confident and wrong. If the detail matters, the picture has to come from somewhere that knows it.
Checkpoint
Change one element per attempt, save the version you liked, and translate abstract ideas into things that can actually be drawn.
๐ Quiz
Question 1 of 2Your brief is detailed but the result still feels generic. Which slot is most often the one left empty?