Subject, scene, style
- ai images
- prompting
A strong image prompt fills four slots: subject (what is in frame), scene (what surrounds it), style (how it should look) and light (time of day, mood, direction). Naming a thing — “a dog” — leaves three of four slots to chance. Describing all four gives the model independent decisions to make well, and gives you independent dials to turn when you iterate.
The four slots
- Subject — one clear subject beats three competing ones. Add the telling details: what it is doing, what state it is in.
- Scene — where it sits, what is around it, how busy or calm the setting is.
- Style — a look, not praise: “quiet editorial photography”, “flat vector illustration”, “soft watercolor with visible paper grain”.
- Light — direction and mood: “soft morning light through sheer curtains”, “hard noon sun, crisp shadows”, “blue hour, city glow”.
Style words guide when they describe something the model can render. Generative systems do not reward praise: “stunning” and “amazing” decide nothing.
Weak prompt
a nice coffee photo, make it amazing
Stronger prompt
One ceramic cup of coffee with gentle steam (subject), on a wooden windowsill with a small plant (scene), calm editorial photography with shallow depth of field (style), soft morning light through sheer curtains (light)
Why the stronger one works
- Each of the four slots answers a question the model would otherwise guess.
- “Calm editorial photography” describes a renderable look; “amazing” describes a feeling no model can act on.
- The result is iterable: dislike the light? Change only the light slot.
A bad example
Ask: “A dog.”
Every meaningful decision — setting, style, time of day, mood — is made by the model. You get an image, but not yours.
A better example
Ask: “A corgi asleep on a windowsill (subject), minimalist room with two plants (scene), soft watercolor illustration with visible paper grain (style), warm late-afternoon light (light).”
Same model, same effort — four deliberate choices instead of none.
Why it works
The four slots are separable decisions. Filling them all makes the first draft close enough to iterate on, and when something is off you can tell which slot caused it. That is how prompt writing becomes engineering instead of guessing.
Practice
Build your first image prompt
You need a header image for an article about slow mornings: a cup of coffee by a window, calm and warm, nothing busy in the frame.
Fill all four slots — subject, scene, style and light — then compare with the reference version.
A strong example (self-assessment — not machine-graded)
Subject: one ceramic cup of coffee with gentle steam rising. Scene: a wooden windowsill with a small plant, quiet and uncluttered. Style: calm editorial photography with shallow depth of field. Light: soft morning light through sheer curtains, warm and low-contrast.
Hint
One clear subject beats three competing ones.
Style words work when they describe a look — “quiet editorial photography” guides; “amazing” does not.
Why this is the answer
The four slots — subject, scene, style, light — are the difference between naming a thing and describing an image. Each slot gives the model an independent decision to make well, and each one you skip is a decision made for you. Iteration starts from a complete first draft; that is what makes it possible to fix one thing at a time.
Transfer
- Slides and headers: the four slots map one-to-one onto a usable brief.
- Iteration: when something looks wrong, fix the slot responsible — not the whole prompt.
- Briefs for others: the four slots are also a checklist to review someone else’s prompt.
Key takeaways
- Subject, scene, style, light — four slots, one recipe.
- Style words describe a look; praise words do nothing.
- A complete first draft exists to be iterated, one slot at a time.
Next
Next: Composition and framing — where things sit in the frame. The generative AI entry covers how these systems construct images in the first place.