Compare these two prompts:
"Create a picture of a person working."
vs.
"A realistic editorial photo of a startup founder working alone at 7 AM, viewed from slightly behind, with a laptop in the foreground and soft morning light entering through a window."
Same subject. Completely different output quality. The first one leaves every visual decision up to the model's defaults — which means generic. The second one reads like a shot list a photographer would actually get.
The difference isn't prompt length. It's how many of the right decisions you've made before you hit generate. Most people write longer prompts when an image comes out wrong. What actually fixes it is controlling six specific variables.
The 6-Step Visual Direction Formula
Run through these before you generate anything:
- What am I showing? — Person, product, space, concept, or process.
- What should people notice first? — Make the focal element visually dominant.
- Where should everything sit? — Composition and visual weight.
- What should it feel like? — Mood, set through lighting and color palette.
- What makes it look real? — Materials, texture, and natural imperfection.
- Where is it going to live? — Mobile feed, header graphic, thumbnail, card.
Skip any of these and the model fills the gap with whatever's statistically average for that subject — which is exactly what produces the flat, overly-symmetrical, slightly plastic look most AI images have.
Prompt Templates by Variable
1. Composition & Negative Space
If the image needs room for a headline or text overlay, say so directly — don't leave it to chance.
Side-weighted hero (subject pushed to one side, space left for text):
Centered focal point (product shots, hero images):
2. Camera Angle & Scale
Angle changes the emotional register of a shot more than almost any other variable.
Low angle — reads as authority, presence:
Desk-level — reads as intimate, relatable, "in the room":
3. Lighting & Palette
Lighting builds structure. Color carries emotional tone. Skip vague terms like "HD lighting" — name the source and direction instead.
Soft window light:
High-contrast side light:
4. Fixing the "Plastic AI" Look
This is the single most common complaint about AI images, and it's fixable with prompt language, not just post-processing. Tell the model to respect real materials and imperfection instead of defaulting to airbrushed smoothness.
Authentic, unposed portrait:
5. Social & Thumbnail Framing
Anything headed for a mobile feed needs to read clearly at a fraction of its full size — that means one dominant visual idea, not a composition full of detail.
Thumbnail-first clarity:
Refine One Variable at a Time
When a generated image is close but not right, don't rewrite the whole prompt. Change one thing and regenerate — that's the only way to learn which instruction is actually responsible for which result.
- Too dark? Touch only the lighting line.
- Subject too cramped? Adjust framing or camera distance — nothing else.
- Looks artificial? Add material and texture language, switch to documentary/unposed framing.
- No room for text? Shift subject placement and explicitly declare negative space.
Change everything at once and you'll never know what fixed it. Change one variable and you build an actual mental model of how the tool responds — which is what separates people who can direct AI images reliably from people who are still rerolling and hoping.
Universal Template
Cover / Featured Image Prompt:
