PROMPT ENGINEERING

How to Write AI Image Prompts (With Before-and-After Examples)

2 min read · Last reviewed 2026-09-30

AI image tools are impressive, but they can't read your mind. Type "a coffee shop" and you'll get a generic coffee shop that looks like a stock photo from nowhere in particular. The fix is simple: describe the image as if you were briefing a photographer.

The five-part checklist

Almost every strong image prompt covers five things. You don't need all five every time, but the more you include, the less the tool has to guess.

Subject: what is in the picture, specifically. Not "a dog," but "an old golden retriever with a grey muzzle."

Setting: where it is and when. "On a porch in the rain at dusk" does a lot of work in a few words.

Style: what kind of image it is. A photograph, a watercolour, a flat vector illustration, a 3D render, an anime still. This one choice changes the result more than anything else.

Mood and lighting: warm, moody, bright, cinematic, soft morning light, harsh neon. Lighting is what makes an image feel professional.

Framing: close-up, wide shot, from above, eye level. Think about where the camera is.

Before and after: a product shot

Weak: "a candle on a table."

Strong: "Product photograph of a white soy candle in an amber glass jar on a light oak table, soft window light from the left, shallow depth of field, calm and minimal, eye-level close-up."

The strong version tells the tool the style (product photo), the details (amber jar, oak table), the lighting, the mood, and the framing. The result looks like something you could put on a store page.

Before and after: a social media graphic

Weak: "a robot learning."

Strong: "Anime-style illustration of a white humanoid robot apprentice studying glowing holographic code in a traditional Japanese dojo at night, lanterns and cyan light, dramatic and hopeful, medium shot."

Common mistakes

Writing a wish list instead of a scene. Stuffing in fifteen unrelated ideas confuses most tools. Pick one clear scene.

Expecting perfect text. Many image tools still struggle with words inside the image, especially long ones. If you need exact text, add it afterward in a design tool like Canva.

Using living artists' names as a shortcut. Beyond the ethical questions, describing the style you want ("bold ink lines, muted earth tones") gives you more control anyway.

Iterate like a photographer

Treat each image as a test shot. Change one thing at a time: the lighting, then the angle, then the style. If you change everything at once, you won't know what made it better.

The honest bottom line

Different image tools respond to prompts differently, and they change often. The five-part checklist works across all of them because it's really just clear description. Start there, then learn the quirks of whichever tool you settle on.