How to Write Better AI Image Prompts (With 50 Examples)

Most people who complain that AI image generators are "all the same" are prompting them the same way.

"A dog in a park." "A futuristic city." "A woman working at a laptop."

You get the obvious interpretation of the obvious prompt. Generic, forgettable, useful for nothing specific.

The gap between a beginner and an experienced AI image creator isn't about access to better tools. It's about understanding that these tools aren't search engines -- they're systems that respond to language with image logic. Once you understand that logic, you can steer them precisely.

I've been doing creative work for 25 years and have spent the past 18 months learning this particular skill. Here's everything that actually works.

The Anatomy of a Good Prompt

Before the 50 examples, the framework. A strong AI image prompt typically has five components -- you don't always need all five, but when something's coming out wrong, it's usually because one of these is missing.

1. Subject -- What's the main thing in the image?

2. Style/Aesthetic -- What visual language should it use?

3. Lighting -- How is the scene lit?

4. Composition/Camera -- Where are we looking from? How is it framed?

5. Technical quality modifiers -- Rendering quality, detail level, mood.

The beginner prompt hits #1 and sometimes #2. The professional prompt hits all five.

Bad: A coffee shop

Better: A small independent coffee shop, early morning, golden hour light filtering through steam, shot from a low angle, warm and intimate, film photography aesthetic, 35mm grain

The second prompt is doing five things at once. Every word is doing work.

---

Part 1: Subject and Composition

Describe the Subject with Specificity

The more specific you are about the subject, the more distinctive the result. Generic nouns give you average interpretations. Specific descriptions give you something unique.

Example prompts:

1. A 1950s diner waitress with dark red lipstick, holding a coffee pot, caught mid-laugh
2. A mechanical engineer examining the inside of a turbine, safety glasses pushed up on her forehead, fluorescent workshop lighting
3. An elderly fisherman repairing nets on a wooden dock at dusk, worn hands, weathered face
4. A young child discovering a firefly for the first time, wonder in their eyes, suburban backyard, late summer
5. A startup founder staring at a laptop screen at 2am, empty coffee cups on the desk, city lights through the window behind

Notice what each prompt does: it gives the subject a specific action, a specific emotion, a specific setting detail. Not just "a woman working." A founder, 2am, empty coffee cups.

Control Composition Deliberately

Composition is one of the most overlooked elements. You can tell the AI where to put things.

Example prompts:

1. Shot from above looking down, a farmer's market with colorful produce arranged in geometric patterns, birds-eye view, aerial perspective
2. Close-up portrait, extreme shallow depth of field, a chef tasting from a wooden spoon, bokeh kitchen background, nose and lips sharp, eyes slightly soft
3. Wide angle interior shot, a mid-century modern living room, fireplace on left, floor-to-ceiling windows on right, couple seated facing away from camera
4. Dutch angle composition, a city street at night, rain-slicked asphalt, noir atmosphere, tension
5. Symmetrical composition, a long hotel hallway, ornate carpet runner, warm sconces, perspective vanishing point, The Shining inspired but elegant

---

Part 2: Style and Aesthetic

Reference Artistic Movements

Trained on enormous libraries of art, these models understand art historical terms better than most people do. Use them.

Example prompts:

1. Bauhaus design poster, geometric shapes, primary colors, bold sans-serif typography, 1920s graphic design aesthetic
2. Art nouveau illustration, flowing organic lines, a woman with long flowing hair intertwined with botanical elements, Alphonse Mucha inspired
3. Brutalist architecture, concrete tower block, dramatic low angle, stormy sky, mid-century urban photography
4. Ukiyo-e woodblock print style, a mountain landscape with dramatic waves in the foreground, Hokusai inspired
5. Memphis design aesthetic, squiggly lines, pastel colors, 1980s Italian design, playful geometric shapes

Reference Photography Styles

For photorealistic images, photography terminology is your best friend.

Example prompts:

1. Documentary photography, black and white, a busy subway platform during rush hour, natural light, gritty, high contrast
2. Editorial fashion photography, Vogue-style, minimalist white background, woman in structural architectural clothing, harsh shadows
3. National Geographic-style wildlife photography, a tiger partially hidden in tall grass, telephoto lens, golden hour side light
4. Product photography, white seamless backdrop, soft box lighting, premium skincare bottle, clean and clinical
5. Reportage photography, candid street scene, a food vendor on a rainy street in Tokyo, steam rising, neon reflections

Reference Film and Visual Media

Referencing films or TV shows gives the AI a complete visual language to draw from.

Example prompts:

1. Wes Anderson color palette, pastel pinks and yellows, symmetrical composition, a small town train station, whimsical, deadpan
2. Blade Runner 2049 aesthetic, massive scale, desolate landscape, warm amber light in a grey world, lone figure walking
3. Studio Ghibli inspired illustration, lush fantasy forest, small wooden house with smoke from the chimney, soft watercolor feel
4. 1970s Eastmancolor film look, warm orange tones, a suburban American street, nostalgic, slightly faded
5. Wong Kar-wai aesthetic, blurred motion, neon-lit Hong Kong alleyway, melancholy, rain-soaked

---

Part 3: Lighting

Lighting changes everything. Same subject, different lighting = completely different image.

Natural Lighting

Example prompts:

1. Golden hour backlight, a silhouette of a couple on a beach, warm orange glow, long shadows, romantic
2. Overcast diffused light, a forest path covered in autumn leaves, soft, even, no harsh shadows, melancholy
3. Harsh midday sun, a farmer working in a field, high contrast shadows, baked earth, physical labor
4. Blue hour just before dawn, a city skyline reflected in a calm river, deep blue tones, tranquil, professional architectural photography
5. Dramatic storm light, a lighthouse on rocky cliffs, dark clouds with a single beam of light breaking through, Turner painting atmosphere

Artificial Lighting

Example prompts:

1. Single practical lamp, a writer at a wooden desk late at night, pool of warm light on the desk, everything else in shadow, intimate
2. Neon sign light, a bartender at an empty bar after closing, blue and pink neon reflections on wet surfaces, noir
3. Candlelight only, a Victorian dinner table set for many guests, nobody present, long shadows, gothic atmosphere
4. Film noir lighting, low key, a detective in a trench coat, venetian blind shadows striping across his face, black and white
5. Medical fluorescent lighting, harsh, blue-white, a surgeon scrubbing in, sterile, clinical precision

---

Part 4: Technical Modifiers That Actually Work

These are the additions at the end of prompts that tell the AI about rendering quality and style.

For Midjourney

1. --ar 16:9 -- widescreen (great for banners and editorial)
2. --ar 4:5 -- portrait (ideal for Instagram feed posts)
3. --ar 1:1 -- square (social media standard)
4. --style raw -- less Midjourney "processing," more photographic accuracy
5. --stylize 0 -- minimal artistic interpretation, closer to literal prompt
6. --stylize 750 (or higher) -- maximum artistic interpretation, distinctive Midjourney look
7. hyperrealistic, 8k, sharp focus, ultra detail -- for photorealistic renders
8. painterly, loose brushwork, impasto texture -- for painterly illustration feel

For DALL-E 3.5

1. Describe exactly what you want, step by step -- DALL-E responds better to explicit instructions than to mood words
2. Do not include text or words in the image -- stops DALL-E from adding random text
3. The image should look like a real photograph, not an illustration -- clarifies output type when needed

For Ideogram

1. Use quotation marks around text you want to appear in the image: A coffee shop sign reading "OPEN" -- Ideogram reads this literally and renders the text
2. bold typography, sans-serif, clean layout -- Ideogram excels at type-forward design
3. poster design, large central text, minimal illustration -- plays to Ideogram's text-first strengths

---

Part 5: The Prompts I Use Most in Client Work

These are the actual prompt structures I return to when I need reliable, professional results on client projects.

For brand photography:

1. [Product description] on [surface/setting], professional product photography, soft studio lighting, clean background, commercial grade, subtle shadows

Swap in your product. Adjust surface and setting for the brand's aesthetic. This produces usable results 70% of the time without additional iteration.

For editorial illustration:
[Scene description], editorial illustration, [specific magazine] aesthetic, [color palette], hand-drawn feel, [era if relevant]

For social media creative:
[Key visual element], flat design, minimal, bold [brand color] and [secondary color], clean white space, modern graphic design

For concept mood boards:
[Mood word] [setting], [lighting type], [film reference or artist reference], high quality, detailed, --ar 16:9

---

The Iteration Mindset

Here's the thing people miss: prompting isn't a single-shot game. It's a conversation.

Your first generation is a direction, not a destination. Look at what came out. Identify what's wrong. Adjust one or two variables at a time. Generate again.

The users who get the best results from these tools aren't the ones with the best first prompts. They're the ones who iterate quickly and read their outputs like feedback rather than failures.

I'll typically run 5-15 generations for any concept I'm serious about. The first prompt establishes the territory. The last few are refinements on something that's already close to what I want.

A few iteration strategies that work:

The "plus" prompt: Take your current prompt and add one specific qualifier you think it's missing. If the mood is wrong, add a mood descriptor. If the composition is off, add a camera instruction. Don't rewrite everything.

The "extract and remix": When Midjourney produces one image in a grid that's close to right, upscale it and use the vary region tool to fix specific elements while preserving what's working.

The "describe and steal": Midjourney's /describe command generates a prompt from any image you upload. When you see an image with the look you want, describe it -- Midjourney will reverse-engineer what it thinks the prompt was. Use that as your starting point.

---

Common Mistakes (and How to Fix Them)

Too many competing ideas. "A futuristic cyberpunk forest with waterfalls and ancient ruins and anime characters and glowing neon." The AI has to make trade-offs on every conflicting element. Focus on one thing and describe it precisely.

Forgetting lighting entirely. Most prompts skip lighting and then wonder why the image feels flat. "Warm golden hour side light" or "dramatic overhead soft box" transforms the same subject completely.

Using low-specificity style terms. "Beautiful" doesn't mean anything to a model. "Beautiful in the style of Edward Hopper -- isolated, quiet, ordinary American life, oblique light through windows" means something. The more specific your style references, the more distinctive the output.

Not using aspect ratios. Default square or slightly landscape outputs won't work for most professional uses. Add --ar 16:9 for widescreen, --ar 4:5 for social feeds, --ar 2:3 for vertical editorial layouts. Learn the ratios your work actually needs.

Prompting for what you don't want. "No text, no watermarks, not blurry" -- negative prompts work better as explicit negative weights in Midjourney (--no text) or as structural parts of the DALL-E prompt. Saying what you don't want is less effective than being specific enough about what you do want.

For a look at how these generators compare against each other, our Midjourney vs DALL-E vs Ideogram comparison covers the technical differences in detail. And if you're using these tools for commercial work, our Midjourney review covers the pricing tiers that make sense for professional volume. For designers who need text in their images, the Ideogram 2.0 review explains what makes it the strongest tool for that specific use case.

The skill of prompting is the skill of saying exactly what you mean with language that a visual system can interpret. That's not so different from what good advertising writers have always done -- the brief determines the output. Write a vague brief, get a vague result. Write a specific one, and you're surprised at how often you get exactly what you asked for.