How to Write AI Image Prompts: The Complete Guide for Every Model (2026)

How to Write AI Image Prompts: The Complete Guide for Every Model

Most people type "a cool dragon" and wonder why the output looks generic. The difference between mediocre AI images and stunning ones isn't the model — it's the prompt. This guide breaks down exactly how image prompts work across every major model, with 15+ examples you can steal and modify. For instant image prompts, try our image prompt generator or model-specific tools for Midjourney.

The Anatomy of a Good Image Prompt

Every effective image prompt contains up to six building blocks. You don't need all six every time, but knowing the full toolkit means you can control the output instead of hoping for something usable.

1. Subject

The core of what you want to see. Be specific. "A dog" gives you a generic stock photo dog. "A weathered border collie with heterochromia, mid-leap over a fallen log" gives you something worth keeping.

Principles:

• Name the subject precisely — species, age, gender, action, expression
• Describe what they're doing, not just what they are
• Include spatial relationships ("standing in front of," "reflected in," "towering above")
• Quantity matters — "two" is very different from "a crowd of"

2. Style

The visual language. This is where most prompts either shine or collapse into "generic AI art."

Major style categories:

3. Lighting

Lighting is the single most underused element in image prompts. Photographers spend years learning this. You can describe it in five words.

Key lighting terms:

• Golden hour — warm, directional, long shadows. The Instagram favorite.
• Blue hour — cool, ambient, moody. Right before sunrise or after sunset.
• Rembrandt lighting — dramatic, triangle shadow on cheek. Portraiture classic.
• Rim/backlighting — subject outlined in light, dramatic silhouette effect.
• Overcast/diffused — soft, even, no harsh shadows. Natural documentary feel.
• Neon/artificial — colored light sources, urban night photography, cinematic.
• Chiaroscuro — extreme contrast between light and dark. Caravaggio territory.
• Volumetric lighting — visible light rays, fog, atmosphere. God rays through forest canopy.
• Studio lighting — controlled, professional, clean. Product shots, headshots.

4. Composition

How the image is framed. Most people skip this entirely, which is why AI defaults to center-framed, eye-level shots.

Composition terms that work:

• Camera angle: bird's eye view, worm's eye view, Dutch angle, over-the-shoulder
• Shot type: extreme close-up, medium shot, wide establishing shot, macro
• Lens effect: fisheye, tilt-shift (miniature effect), bokeh (blurred background), lens flare
• Framing: rule of thirds, symmetrical, leading lines, frame within a frame
• Depth: shallow depth of field, deep focus, foreground/midground/background layering

5. Mood and Atmosphere

The emotional temperature of the image. This is often the difference between a technically correct image and one that feels like something.

Mood descriptors:

• Ethereal, haunting, serene, ominous, whimsical, melancholic, chaotic, intimate
• Nostalgic, surreal, clinical, raw, dreamlike, tense, celebratory, lonely

Atmosphere descriptors:

• Foggy, dusty, smoky, rainy, hazy, crisp, humid, frozen
• Dense, sparse, cluttered, minimal, lush, barren, vibrant, muted

6. Technical Specifications

Model-specific parameters that control output quality and format.

• Aspect ratio: 16:9 (landscape/cinematic), 9:16 (vertical/mobile), 1:1 (square), 4:5 (Instagram portrait)
• Quality terms: highly detailed, 8K, photorealistic, ultra-sharp, high resolution
• Rendering: octane render, unreal engine, ray tracing, V-Ray (for 3D)
• Camera: shot on Hasselblad, 35mm film, Fujifilm color science, Canon 85mm f/1.4

Model-Specific Tips

Each model interprets prompts differently. Here's what works (and what doesn't) for each.

DALL-E 3 (ChatGPT)

DALL-E 3 is integrated into ChatGPT, which means you write natural language and ChatGPT actually rewrites your prompt before sending it to the image model. This is both a strength and a limitation.

What works well:

• Natural, conversational descriptions
• Complex scenes with multiple elements and spatial relationships
• Text in images (DALL-E 3 handles text better than most)
• Specific art styles and cultural references

Tips:

• Be descriptive but don't keyword-stuff — write like you're describing the image to someone
• If ChatGPT's rewrite isn't producing what you want, ask it to use your exact prompt
• For text in images, put the text in quotes and keep it short (under 6 words works best)
• Specify what you DON'T want in natural language ("without any text overlay," "no people in the background")

For 50 copy-paste DALL-E prompts, see our ChatGPT image prompts guide.

Example prompt:

A cozy independent bookshop interior, late afternoon. Warm golden light streaming through tall windows, casting long shadows across wooden shelves overflowing with books. A ginger cat sleeping on a stack of hardcovers near the register. Dust motes visible in the light beams. Shot from a low angle, looking up at the shelves. Watercolor illustration style with loose, expressive brushwork and a warm ochre and sage color palette.

Midjourney

Midjourney excels at aesthetic quality and tends to produce "beautiful" images by default. It responds well to concise, keyword-driven prompts and has its own parameter system.

What works well:

• Shorter prompts (Midjourney often performs better with less text)
• Stylistic keywords and aesthetic references
• Parameters like --ar, --style, --chaos, --stylize
• Photography terms and camera references

Key parameters:

• --ar 16:9 — aspect ratio
• --s 250 — stylize level (0-1000, higher = more Midjourney aesthetic)
• --c 30 — chaos (0-100, higher = more variation)
• --no text, borders — negative prompts
• --style raw — less Midjourney "beautification," more literal

Tips:

• Front-load the most important elements
• Use :: for weighted sections — portrait of a woman::2 neon cityscape background::1 emphasizes the portrait
• Keep prompts under 60 words for best results
• Use --style raw when you want the model to follow your prompt literally

For a complete Midjourney guide with advanced techniques, read our Midjourney v7 prompting guide.

Example prompt:

weathered lighthouse on a rocky cliff, stormy ocean, dramatic clouds breaking to reveal sunset, cinematic lighting, shot on medium format film, moody atmosphere --ar 16:9 --s 400

Stable Diffusion (and SDXL)

Stable Diffusion gives you the most control but requires the most prompt engineering. It's an open-source model that runs locally or through services like Stability AI, Automatic1111, or ComfyUI.

What works well:

• Keyword-heavy, comma-separated prompts
• Quality boosters ("masterpiece, best quality, highly detailed")
• Negative prompts (essential — often more important than the positive prompt)
• LoRA and model-specific syntax
• Weighted tokens using (parentheses) for emphasis

Tips:

• Start with quality keywords: "masterpiece, best quality, highly detailed, sharp focus"
• Use negative prompts aggressively: "low quality, blurry, deformed, watermark, text, bad anatomy, extra fingers"
• Parentheses increase weight: (golden hour lighting:1.3) emphasizes the lighting
• Order matters — put the most important elements first
• Checkpoint model choice matters more than in other platforms — a photorealistic checkpoint handles photos better than an anime checkpoint

Negative prompt template (use as a starting base):

(worst quality:1.4), (low quality:1.4), (normal quality:1.4), blurry, bad anatomy, bad hands, extra digits, fewer digits, cropped, watermark, text, signature, jpeg artifacts, deformed, disfigured, mutation, extra limbs

Example positive prompt:

masterpiece, best quality, a Japanese tea house in autumn, maple leaves in brilliant red and orange, koi pond reflecting the trees, soft morning mist, traditional architecture, highly detailed, warm color palette, Studio Ghibli aesthetic, (volumetric lighting:1.2), serene atmosphere

Flux (by Black Forest Labs)

Flux is the newer model gaining serious traction for its prompt adherence and photorealistic capabilities. Available in Flux Pro, Flux Dev, and Flux Schnell variants.

What works well:

• Detailed, natural language descriptions (similar to DALL-E 3)
• Precise spatial relationships and complex compositions
• Photorealistic outputs with accurate human anatomy
• Text rendering (among the best of any model)
• Consistent style across multiple generations

Tips:

• Flux follows long, detailed prompts accurately — don't fear length
• Natural language outperforms keyword-stuffing (unlike Stable Diffusion)
• Specify camera and lens for photorealistic shots
• For maximum control, describe foreground, midground, and background separately
• Flux handles multiple subjects and their relationships better than most models

For Flux-specific techniques and advanced prompts, see our Flux Pro prompting guide.

Example prompt:

A photorealistic aerial view of a coastal Italian village at golden hour. Terracotta rooftops cascade down a hillside toward a small harbor with colorful fishing boats. Narrow cobblestone streets are visible between buildings. Mediterranean cypress trees dot the hillside. Warm sunset light bathes the scene in amber, with long shadows stretching east. Shot on a DJI Mavic with a wide-angle lens, slight tilt-shift effect making the village look miniature. 16:9 aspect ratio.

15 Example Prompts With Explanations

Here are prompts across different use cases, with annotations explaining why each element works.

Product Photography

Prompt 1 — Minimalist Product Shot:

A frosted glass perfume bottle on a smooth marble surface, soft diffused studio lighting from above, water droplets on the glass, a single orchid petal in the foreground slightly out of focus, clean white background, product photography, shot on Phase One IQ4 150MP, 4:5 aspect ratio

Why it works: Specifies material (frosted glass), surface (marble), lighting (diffused, from above), details (water droplets), depth (orchid petal out of focus), and technical specs (camera, aspect ratio). Every element serves the "premium product" goal.

Prompt 2 — Lifestyle Product Context:

Flat lay of a leather journal, brass pen, and espresso in a ceramic cup on a worn oak desk, morning sunlight from a window on the left creating soft shadows, a smartphone showing a calendar app slightly cropped at the edge, warm earth tones, editorial photography style for a lifestyle brand

Portraits

Prompt 3 — Environmental Portrait:

An elderly Vietnamese fisherman mending nets on a wooden boat at dawn, Hoi An river, golden hour sidelight illuminating weathered hands, shallow depth of field with lanterns blurred in the background, National Geographic documentary style, shot on 85mm f/1.4, warm color grade

Why it works: Specific subject (not "old man"), specific location, specific time, specific action, specific lens. The "National Geographic documentary style" anchors the entire aesthetic.

Prompt 4 — Stylized Fashion Portrait:

Editorial fashion portrait, a model with shaved head and geometric face paint in cobalt blue, wearing an oversized deconstructed blazer, standing in a concrete brutalist parking garage, hard directional flash creating sharp shadows, high contrast black and white with only the face paint in color, Helmut Newton meets Hype Williams, 3:4 aspect ratio

Landscapes

Prompt 5 — Natural Landscape:

A glacial lake in Patagonia at blue hour, mirror-still water reflecting jagged snow-capped mountains, single dead tree in the foreground breaking the symmetry, thin clouds streaking across a deep blue sky, long exposure effect on the water, shot on large format film, Ansel Adams tonal range, panoramic 21:9 aspect ratio

Prompt 6 — Urban Landscape:

Tokyo alleyway at night in the rain, neon signs in Japanese reflecting off wet asphalt, a single figure with a transparent umbrella walking away from camera, steam rising from a ramen shop vent, cyberpunk color palette of magenta and teal, shallow depth of field, street photography, Liam Wong inspired

Illustration Styles

Prompt 7 — Vintage Botanical:

Scientific botanical illustration of a Venus flytrap (Dionaea muscipula) in various stages — closed trap, open trap, and catching a fly. Detailed cross-section showing trigger hairs and digestive glands. Aged parchment background, hand-drawn with fine ink line work and delicate watercolor tinting, 18th century natural history style, labeled in elegant cursive handwriting

Prompt 8 — Children's Book:

A bear cub and a fox kit building a treehouse together in an autumn forest, the bear holding a plank while the fox hammers a nail, scattered tools and acorns around the base of the oak tree, golden afternoon light filtering through red and orange leaves, warm and inviting children's book illustration style, gouache on paper texture, slightly flat perspective, inspired by Jon Klassen

Prompt 9 — Concept Art:

A floating marketplace on a terraformed Mars, domed glass structures connected by covered bridges over red dust plains, vendors selling hydroponic produce, mixed human and android shoppers, two pale moons visible in a pinkish sky, hard science fiction aesthetic, detailed concept art for a film production, wide establishing shot, Greg Rutkowski lighting

Abstract and Artistic

Prompt 10 — Abstract:

Abstract fluid art composition, deep ocean blues and liquid gold merging and separating in organic patterns, microscopic detail in the paint cell structures, high contrast between dark prussian blue depths and bright metallic gold peaks, macro photography of actual fluid pour, dramatic studio lighting from the side creating depth in the paint surface, 1:1 square format

Prompt 11 — Surrealist:

A massive antique pocket watch melting over the edge of a cliff overlooking the ocean, the clock face filled with water and small fish swimming among the gears, surrealist painting, muted earth tones with a single vivid turquoise ocean, soft sfumato edges, Dalí meets Magritte, oil on canvas texture, museum quality

Architecture and Interior

Prompt 12 — Architectural Visualization:

A modern Japanese-Scandinavian fusion house, floor-to-ceiling windows framing a bamboo garden, exposed timber ceiling beams, polished concrete floors with a sunken living area, minimal furniture in natural oak and bouclé fabric, late afternoon sun creating warm rectangles of light on the floor, architectural photography by Hiroshi Sugimoto, wide angle, clean and serene

Prompt 13 — Interior Design:

A maximalist home library, floor-to-ceiling bookshelves on every wall, rolling ladder in aged brass, a tufted emerald velvet reading chair with a floor lamp, stacks of books on a Persian rug, warm incandescent lighting, afternoon light through a single arched window creating god rays in the dusty air, overhead view looking down at the chair, rich and moody color palette

Food Photography

Prompt 14 — Overhead Food Shot:

Overhead flat lay of a homemade pasta dinner spread, hand-cut pappardelle with wild mushroom ragu in a rustic ceramic bowl, torn crusty bread, half-empty wine glass casting a red shadow, scattered parmesan shavings and fresh thyme, worn wooden table surface with flour dusting, moody food photography, natural window light from the top, dark and warm color grade, shot on Fuji GFX 100S

Graphic Design and Branding

Prompt 15 — Logo Concept:

A minimal, geometric logo mark for a sustainable architecture firm called "Verdant." The mark should merge a leaf shape with architectural blueprint lines, using clean vectors with no gradients. Single color — deep forest green (#2D5016) on white. Scalable from favicon to billboard. Modern, clean, and immediately readable. Show the mark alone, no text, centered on white background, vector illustration style

Common Mistakes (and How to Fix Them)

Mistake 1: Being Too Vague

"A beautiful landscape" tells the model nothing. Beautiful how? Landscape where? What season? What time of day?

Mistake 2: Keyword Stuffing Without Structure

"Beautiful amazing stunning gorgeous incredible breathtaking epic masterpiece" is wasted tokens. These words all mean roughly the same thing to the model. Use specific descriptors instead of stacking superlatives.

Mistake 3: Ignoring Negative Prompts

Especially for Stable Diffusion: if you don't specify what you don't want, the model will give you its default — which often includes watermarks, extra fingers, blurry backgrounds, and text overlays. Always use negative prompts for SD/SDXL.

Mistake 4: Fighting the Model's Strengths

Midjourney makes everything beautiful. If you want raw, ugly, or documentary-style, use --style raw or switch to a model better suited to realism. DALL-E 3 handles text well. SD gives you technical control. Flux does photorealism. Use the right tool.

Mistake 5: Forgetting Aspect Ratio

The default is usually square (1:1). If you want a landscape scene, specify 16:9 or wider. If you want a phone wallpaper, specify 9:16. If you want an Instagram portrait, specify 4:5. Aspect ratio dramatically affects composition.

Mistake 6: Not Iterating

Your first prompt is a draft. Generate, evaluate, adjust. Move the lighting. Change the camera angle. Add a foreground element. Remove a detail that's cluttering the composition. The best images come from 3-5 iterations, not one lucky shot.

Advanced Techniques

Multi-Subject Composition Control

When your image has multiple subjects, describe their spatial relationship explicitly:

In the foreground: a child's hand reaching up to release a red balloon. In the midground: the balloon rising past the branches of a cherry blossom tree. In the background: the balloon as a tiny dot against a vast cloudy sky. Shot from below looking up, vertical composition, each layer at a different focus distance.

Style Mixing

Combine two unexpected styles for unique results:

Art Deco poster design meets Japanese woodblock print. A great wave (Hokusai-inspired) rendered in geometric Art Deco patterns, gold and deep navy color palette, strong linework, limited color print aesthetic, vintage poster format with aged paper texture

Time and Sequence

Describe temporal elements for dynamic images:

A coffee cup on a desk, time-lapse style showing the coffee at three stages: full and steaming on the left, half-empty in the center, empty with lipstick stain on the right. Same desk, same angle, changing natural light from morning (cool) to afternoon (warm). Triptych layout, editorial photography.

Texture and Material Focus

When material matters more than subject:

Extreme macro photography of rust patterns on an abandoned ship hull. Iron oxide in deep oranges and browns, flaking paint revealing layers of color underneath — red, then green, then bare metal. Water droplet trapped in a rust crater reflecting the overcast sky. Shot with extension tubes, razor-thin depth of field, every metallic crystal visible.

Negative Space and Minimalism

Less is more — but you have to describe the "less":

A single red maple leaf resting on a vast, smooth grey stone surface, positioned in the lower right third of the frame. The remaining 80% of the image is empty stone texture with subtle grain variation. Minimal, contemplative, Japanese wabi-sabi aesthetic. Muted colors except the leaf. Overhead shot, soft overcast light.

Consistent Character Across Generations

For projects needing the same character in multiple images, create a detailed character brief and reuse it:

Character brief — ALWAYS include in prompts for this character:
"A woman in her mid-30s with short silver-grey hair in a textured pixie cut, warm brown skin, prominent cheekbones, small scar above left eyebrow, wearing a navy blue mechanic's jumpsuit with rolled sleeves, oil-stained hands, confident posture."

Scene-specific prompt: [CHARACTER BRIEF] + standing in a neon-lit garage, leaning against a vintage motorcycle she's restoring, warm tungsten lighting mixing with cool neon, cinematic medium shot, shallow depth of field

The Prompt Formula Cheat Sheet

When you're staring at a blank prompt field, start with this formula and fill in what matters:

[SUBJECT] + [ACTION/POSE] + [SETTING/ENVIRONMENT] + [STYLE/MEDIUM] + [LIGHTING] + [COMPOSITION/CAMERA] + [MOOD] + [TECHNICAL SPECS]

You don't need every element. A strong subject + style + lighting prompt often beats a prompt with all eight elements but weak specifics. Quality of description beats quantity of keywords.

Build Prompts Faster

Writing image prompts from scratch every time is slow. These tools speed up the process:

• Image Prompt Generator — Describe what you want in plain English, get a structured image prompt for any model
• Midjourney Prompt Generator — Build Midjourney-optimized prompts with parameters and style controls
• SurePrompts Builder — Save your best image prompt structures as reusable templates with customizable fields

Keep Reading

• ChatGPT Image Prompts: 50 Copy-Paste Templates — Ready-to-use prompts for DALL-E 3 via ChatGPT
• Midjourney v7 Prompting Guide — Deep dive into Midjourney's latest model
• Flux Pro Prompting Guide — Master prompting for Black Forest Labs' Flux
• Prompt Engineering Basics — Fundamentals that apply to both text and image prompts
• AI Prompt Frameworks — Structured approaches to writing any kind of prompt
• Prompt Templates — Browse 320+ templates including image generation