How to Write AI Image Prompts: A Practical Guide

Most disappointing AI images have the same root cause: a prompt that describes a topic instead of a picture. “A dog” is a topic. “A golden retriever puppy sprinting through shallow surf at golden hour, water droplets frozen mid-air, low-angle action shot” is a picture. This guide gives you a repeatable formula for the second kind — and every example can be tested free in the HelloGen generator.

The 5-part prompt formula

Strong prompts tend to answer five questions, in roughly this order:

1. Subject — who or what, with one or two concrete details: “an elderly clockmaker with wire glasses”, not “a man”.
2. Action or pose — what the subject is doing: “leaning over a workbench, adjusting a tiny gear”.
3. Setting — where and when: “in a cluttered workshop at night”.
4. Style — the visual language: “warm cinematic photography” or “flat vector illustration” or “oil painting”.
5. Light and mood — the finishing layer: “single desk lamp, amber glow, quiet focus”.

You don’t need all five parts every time — but when a result feels generic, it’s almost always because one of these five is missing.

Name a style, or the model picks one for you

“Style” is the highest-leverage word count in your prompt. The same subject becomes a completely different image with different style keywords:

• Photorealism: “cinematic photography, 35mm, shallow depth of field, natural light”
• Flat / vector: “flat vector illustration, geometric shapes, pastel palette, subtle grain”
• Painterly: “oil painting, visible brushstrokes, chiaroscuro lighting”
• Retro print: “vintage travel poster, halftone texture, two-color print”

If you want a matching set of images — say, for a slide deck — reuse the exact same style phrase across all prompts and only change the subject. Our flat illustration collection is built this way.

Be specific about what matters, silent about what doesn’t

Long prompts aren’t automatically better. Every extra clause competes for the model’s attention, so spend words on what you actually care about. If the background is irrelevant, say “clean background” and stop. If color is the point, name exact colors: “terracotta and sage green”, not “nice colors”.

Aspect ratio is part of the prompt

Composition changes with the frame. A 9:16 portrait pushes the model toward close-ups and vertical compositions; 16:9 invites landscapes and scenes. Choose the ratio before generating — HelloGen has 1:1, 4:3, 3:4, 16:9 and 9:16 built in — and mention wide or tall composition in the prompt for extra control.

Iterate like a director, not a slot machine

When a result is close but not right, don’t reroll blindly — change one variable at a time, like a director giving notes:

• Wrong mood? Swap the lighting phrase: “overcast soft light” → “dramatic golden hour”.
• Too busy? Add “minimal composition, negative space”.
• Wrong style era? Anchor it: “1970s film photography” or “modern editorial”.

Because generation on HelloGen is free and takes seconds, three targeted iterations beat thirty random ones — and cost you nothing either way.