The Anatomy of a Great Prompt
The five-part formula behind every great AI image: subject, action, environment, style, and technical specs.
Type a cool dragon into an image model and you will get a dragon. Some dragon. A committee-designed, seen-it-a-thousand-times dragon. The model is not lazy — it is starving. Every word you leave out, it fills in with the statistical average of its training data, and the average of everything is beige. Great prompting is simply refusing to let the model guess.
The five-part formula
Nearly every professional image prompt is built from the same five components, in roughly this order. You will not always need all five — but when a result disappoints, one of these is almost always missing.
- Subject — who or what the image is about: a weathered lighthouse keeper, a chrome espresso machine, a red panda.
- Action — what the subject is doing: mid-leap, pouring latte art, staring down a storm.
- Environment — where and when: a rain-lashed pier at dusk, a sunlit Kyoto alley, a brutalist rooftop.
- Style — the visual language: 35mm documentary photo, Studio Ghibli-inspired illustration, art deco poster.
- Technical specs — the camera and output details: 85mm lens, f/1.8, golden hour, 4K, 16:9.
Before: the prompt everyone writes
a cool dragon
Run it — you will get a generic green-grey dragon on a vague mountain. Not wrong, just nobody's vision. Every missing detail was decided by the average of the internet.
After: the same idea, all five parts
A slender jade-scaled dragon coiled around a crumbling bell tower, exhaling a thin ribbon of silver smoke, in a misty terraced-valley village at dawn, painted in the style of a classical Chinese ink-wash scroll with subtle gold leaf accents, soft diffused light, ultra-detailed, 4:5 portrait
Subject (jade-scaled dragon), action (coiled, exhaling smoke), environment (misty terraced valley at dawn), style (Chinese ink-wash with gold leaf), technical (soft light, 4:5). Same model, different universe.
| Component | Fragment from the prompt above |
|---|---|
| Subject | slender jade-scaled dragon |
| Action | coiled around a bell tower, exhaling silver smoke |
| Environment | misty terraced-valley village at dawn |
| Style | classical Chinese ink-wash scroll, gold leaf accents |
| Technical | soft diffused light, ultra-detailed, 4:5 portrait |
How long should a prompt be? The sweet spot for most models is 25-60 words: enough to specify all five parts, short enough that nothing gets ignored. One vivid sentence beats a paragraph of adjectives. And order matters — describe things in the order of importance, because when a prompt gets crowded, the tail is what the model drops first. We will push each of the five parts much deeper in the next lessons, starting with subject and scene description.
The formula on a product shot
A matte-black wireless earbud case standing upright on a wet slate slab, a single water droplet rolling off its lid, in a dark studio with shallow fog, high-end tech advertisement style, dramatic rim lighting, macro lens, razor-sharp focus, 1:1
Five parts work for anything: portraits, products, landscapes, logos. Swap the subject and keep the skeleton.
The VAR2 Prompt Builder walks you through all five parts with dropdowns — perfect while the formula becomes muscle memory. Open Prompt Builder