Describing Subjects & Scenes Like a Pro
Specificity techniques that turn 'a dog in a park' into an image people stop scrolling for: instances, textures, counts, and layered scenes.
Here is the uncomfortable truth about the subject part of your prompt: the model cannot see what is in your head. 'A dog in a park' contains roughly zero visual information — which dog, of the 340+ breeds? Which park, in which season, on which continent? Professional prompters replace categories with instances, and the difference is instantly visible.
Six specificity techniques
- Name the instance — not 'a dog' but 'a stocky black French bulldog with a grey muzzle and a faded red bandana'.
- Count things — 'three mismatched coffee mugs' renders far more reliably than 'some mugs'.
- Give materials and textures — brushed brass, cracked leather, wet asphalt, hand-thrown ceramic.
- Add age and wear — 'a sun-faded 1970s camper van with rust blooming over the wheel arches' has a life story.
- Place things spatially — foreground, midground, background; left, right, behind, reflected in.
- Pick one focal point — every scene needs a single star; everything else is supporting cast.
Before/after: the portrait upgrade
Portrait of an old fisherman. --vs-- Close-up portrait of a 70-year-old Portuguese fisherman with a deeply lined face, salt-white stubble and cataract-blue eyes, wearing a hand-knitted navy sweater, mending a yellow nylon net with thick weathered fingers, harbor bokeh behind him
Run both halves separately and compare. The second gives the model a specific human being: nationality, age, texture (lined face, stubble), wardrobe, action, and even finger detail.
Building scenes in layers
For anything wider than a portrait, think like a set designer and describe three depth layers explicitly: foreground (what is closest, often slightly out of focus), midground (your focal point lives here), background (context and atmosphere). Models handle layered prompts remarkably well because photo captions in their training data describe scenes exactly this way.
A scene in three layers
Foreground: rain-speckled cafe window with a half-erased chalk menu. Midground: a woman in a mustard-yellow raincoat reading a paperback at a corner table, steam rising from her cup. Background: blurred double-decker buses and umbrellas passing on a wet London street, warm interior light against cold blue dusk outside
Explicit layer labels are perfectly valid prompt text. Notice the color story doing quiet work: mustard against blue dusk.
| Vague (invisible) | Visible upgrade |
|---|---|
| a nice house | a whitewashed Greek island house with cobalt-blue shutters and a bougainvillea-draped gate |
| delicious food | a bubbling margherita pizza, basil leaves curling from oven heat, mozzarella pull mid-slice |
| an interesting character | a wiry courier with mirrored sunglasses, a cracked phone strapped to her forearm |
| dramatic weather | anvil-shaped storm cloud swallowing the horizon, one shaft of sunlight breaking through |
Product scene with counts and materials
Flat-lay of exactly three hand-thrown ceramic espresso cups in matte terracotta, sage and cream, arranged in a triangle on a raw linen cloth, scattered whole coffee beans and one brass spoon, soft window light from the left, top-down view, 1:1
Counting ('exactly three') plus named colors plus materials gives you art-director-level control. Top-down flat-lay is a composition spec — more on that in camera language.
Take one boring noun — dog, house, coffee — and upgrade it with all six techniques. Then generate it. Create an Image