How to Write AI Image Prompts (With a Simple Formula)

A good AI image prompt is not long or clever. It is specific. The model has to decide hundreds of things for every picture: what the subject looks like, where the camera is, what the light is doing. Anything you leave out, it guesses. Better prompts leave fewer guesses.

This guide gives you one formula that works in every major image model, then shows how to adapt it for each one.

The 6-part prompt formula

Write your prompt in this order. You do not need every part every time, but the order matters because models pay the most attention to the start.

1. Subject and action. Who or what, doing what. “An old fisherman mending nets.”
2. Style. Photo, anime, oil painting, 3D cartoon, logo. Pick one.
3. Framing. Close-up, full body, wide angle, top down, low angle.
4. Lighting. Golden hour, soft studio light, neon, candlelight, window light.
5. Mood and colors. Cozy, dramatic, calm. Warm tones, pastel, black and white.
6. Size. Square, vertical 2:3, wide 16:9.

Here is the formula in action.

Before: a fisherman

After: A photorealistic photograph of an old fisherman mending nets on a wooden boat at dawn. Medium shot from the waist up. Soft warm sunrise light from behind, with mist on the water. The mood is calm and quiet. Muted blues and warm highlights. Vertical 2:3 format.

The second prompt is not fancy. It just answers the questions the model would otherwise guess.

Start with the subject, and be concrete

“A woman” gives you an average woman. “A woman in her sixties with silver hair in a loose bun, wearing a green cotton saree” gives you a person. Concrete nouns and details beat adjectives every time.

Swap vague words for things you could point at:

Vague	Concrete
beautiful dress	red silk lehenga with gold zari borders
nice background	a narrow lane with blue doors and hanging plants
cool lighting	pink and blue neon signs reflecting on wet pavement
old car	a 1960s mint-green Ambassador with chrome bumpers

Pick one style and commit

Mixing styles is the most common reason for muddy results. “A photorealistic anime watercolor” asks for three different things. Choose one main style. If you want a blend, name the base style first and the influence second: “an anime illustration with soft watercolor backgrounds”.

Lighting does most of the work

Two prompts with the same subject look completely different under different light. If your images look flat, the fix is usually lighting, not more detail. A few that almost always help:

• Golden hour for warm, flattering portraits and landscapes
• Soft window light from the side for natural indoor photos
• Low-key light with deep shadows for drama
• Backlight with a rim glow to separate the subject from the background

Adapt the prompt to your AI model

The same idea should be written differently for different models. Our AI image prompt generator does this automatically, but here is what changes.

ChatGPT and Gemini

Write plain sentences, like you would brief a photographer. Say what you do not want in words: “No text or watermarks.” If you upload your own photo, add “Keep the face and identity exactly the same as the uploaded photo.”

Midjourney

Use a comma-separated list of the most important words, then parameters at the end:

old fisherman mending nets on a wooden boat, dawn, photorealistic, medium shot, soft sunrise backlight, mist, calm, muted blues --ar 2:3 --v 7 --style raw

--ar sets the aspect ratio, --style raw gives a more photographic look, and --no removes things (--no text, watermark).

Stable Diffusion

Use keywords, give the subject extra weight with brackets, and put unwanted things in the negative prompt box:

(old fisherman mending nets on a wooden boat:1.2), dawn, photorealistic photo, soft backlight, mist, masterpiece, best quality, highly detailed

Negative prompt: blurry, lowres, watermark, text, deformed, bad anatomy, extra fingers

Flux

Flux understands natural language well. Use the ChatGPT style of prompt, but set the aspect ratio in the app instead of the text.

Fix a prompt that is not working

• The subject is wrong: move it to the very start and add two concrete details.
• It looks fake or plastic: add “natural skin texture”, “film grain” or a real lens like “85mm”.
• Too busy: remove half the adjectives and add “clean background” or “negative space”.
• Wrong crop: name the framing (“full body, head to toe visible”) and the aspect ratio.
• Nearly right: keep the prompt and regenerate two or three times before you rewrite it.

A quick checklist

Before you hit generate, check that your prompt answers: what, what style, from where, in what light, how it should feel, and what size. That covers 90 percent of what makes an image good.

Want to see these ideas in real results? Browse the prompt library, where every prompt is shown with the image it made.