Why Prompt Writing Is a Skill Worth Learning
AI image generators respond to language. The more precisely you describe what you want, the more consistently you will get it. This isn't about memorizing magic keywords — it's about developing an intuition for how the model interprets language and using that understanding to communicate your vision clearly.
This guide is structured around the elements that matter most, in roughly the order you should think about them when composing a prompt.
Start With the Subject
The subject is the most important part of your prompt. Be specific. Instead of "a dog", write "a silver husky with blue eyes". Instead of "a city", write "a rain-soaked Tokyo street at midnight, neon reflections on wet asphalt".
Specificity signals confidence to the model. The more concrete your description, the less the model has to guess, and the more your result will match your mental image.
Specify the Medium or Style
AI models have been trained on images spanning centuries of art history and every visual medium imaginable. Naming the medium upfront anchors the entire image in that aesthetic.
Some effective style anchors:
- Photography styles: "cinematic photography", "35mm film photo", "editorial fashion photography", "macro close-up"
- Painting styles: "oil painting", "watercolor illustration", "impressionist painting", "gouache"
- Digital art: "concept art", "matte painting", "digital illustration", "3D render"
- Vintage and retro: "1970s film poster", "vintage Japanese woodblock print", "retro sci-fi pulp illustration"
- Simplified styles: "flat vector art", "pixel art", "minimalist line art", "sticker art"
You can mix styles too. "Watercolor painting with ink linework" or "cinematic photography with painterly textures" can produce striking hybrid results.
Control Lighting to Control Mood
Lighting is one of the most powerful levers in visual communication. It determines mood, drama, and depth more than almost any other element.
- Golden hour: Warm, soft, and romantic. Works beautifully for outdoor portraits and landscapes.
- Dramatic studio lighting: High contrast, sharp shadows. Ideal for powerful portraits and product shots.
- Soft overcast light: Even, diffuse, flattering. Great for portraits where you don't want harsh shadows.
- Neon and artificial: "Neon-lit", "cyberpunk lighting", "rim lighting" — creates vivid, high-energy scenes.
- Candlelight or firelight: Warm, intimate, painterly.
- Volumetric light / god rays: Dramatic beams of light cutting through atmosphere — cinematic and memorable.
Add Composition and Framing
Tell the model how the image should be framed:
- "Close-up portrait" vs "full body shot" vs "wide establishing shot"
- "Bird's eye view" vs "low angle shot"
- "Rule of thirds composition"
- "Symmetrical composition"
- "Depth of field with blurred background" (bokeh)
Use Quality and Detail Boosters
Certain terms consistently improve output quality across most models:
- "Highly detailed", "intricate details", "fine detail"
- "Sharp focus", "crisp", "high resolution"
- "Award-winning", "masterpiece" (use sparingly — it influences style as well as quality)
- "Professional", "studio quality"
Don't overload your prompt with quality boosters. Two or three well-placed terms are more effective than a long list.
Negative Prompts: Telling the Model What NOT to Do
Most AI image generators support negative prompts — a separate field where you list things you want excluded. This is one of the most underused tools available.
Common effective negative prompts:
- "blurry, out of focus, low quality" — prevents soft or degraded results
- "deformed, extra limbs, bad anatomy" — reduces anatomical errors
- "watermark, text, signature" — removes unwanted overlaid elements
- "overexposed, underexposed" — helps with lighting extremes
- "cartoon, anime" — keeps photorealistic prompts from drifting into animated styles (unless that's what you want)
Iterate, Don't Restart
The most important habit in prompt engineering is iteration. Generate a result, identify what's working and what isn't, and adjust one element at a time. This approach builds your understanding of how the model responds and quickly converges on the result you want.
Keep notes on prompts that produce results you like. Over time you will develop a personal library of phrases and structures that consistently work well for your preferred styles.
A Complete Example Prompt
"A silver-haired woman in a flowing red dress standing on a cliff edge at sunset, cinematic photography, dramatic golden hour lighting, shallow depth of field, wide establishing shot, highly detailed, sharp focus, award-winning photography"
Break it down: subject (woman, specific appearance, specific clothing, specific location) + condition (sunset) + medium (cinematic photography) + lighting (golden hour) + technique (shallow depth of field) + composition (wide establishing shot) + quality (highly detailed, sharp focus, award-winning).
This structure gives the model everything it needs. Try it in ImageGen and then begin adjusting individual elements to see how each change affects the output. That iterative practice is what turns prompt writing from guesswork into craft.
A Prompt Library: 20 Ready-to-Use Templates
The fastest way to improve is to study and modify prompts that already work. Here are twenty tested prompts across different use cases:
Portraits
- "Portrait of an elderly Indian woman, deep-set eyes, silver hair, warm amber light from a window, cinematic photography, soft focus background, highly detailed"
- "Young man in a business suit, confident expression, modern office background, editorial photography, dramatic directional lighting, sharp focus"
- "Fantasy warrior woman, braided red hair, battle-worn armour, dramatic backlight, digital concept art, epic fantasy, highly detailed"
Landscapes and Environments
- "Misty mountain valley at sunrise, pine forest, golden rays through clouds, cinematic landscape photography, rule of thirds, breathtaking"
- "Cyberpunk city street at night, neon signs reflected in rain puddles, crowded, atmospheric fog, blade runner aesthetic, cinematic"
- "Ancient temple ruins overgrown with jungle, dappled afternoon light, atmospheric, documentary photography"
Product and Commercial
- "Handcrafted ceramic mug on a wooden surface, morning light, steam rising, cosy coffee aesthetic, product photography, natural light, warm tones"
- "Luxury perfume bottle, black marble surface, dramatic studio lighting, reflections, commercial product photography, sharp focus"
Illustration and Art
- "Children's book illustration, a fox and a rabbit sharing an umbrella in the rain, soft watercolour, warm and friendly, simple background"
- "Art Nouveau poster of a woman reading beneath a tree, flowing botanical border, Alphonse Mucha style, rich colours, detailed"
- "Isometric city scene, tiny buildings and parks, flat graphic style, pastel colours, clean and minimal, digital illustration"
Abstract and Conceptual
- "Abstract representation of artificial intelligence, neural network patterns, glowing nodes, deep blue and electric purple, digital art"
- "Macro photograph of water droplets on a spider web, golden hour light, bokeh background, nature photography, highly detailed"
Advanced Techniques: Going Beyond the Basics
Artist Referencing
Naming specific artists or studios activates the model's knowledge of their distinctive style. "in the style of Wes Anderson" produces a very specific colour palette and symmetrical composition. "Studio Ghibli background art" gives you a particular kind of painterly, atmospheric environment. Use artist references with respect — they guide style, not specific elements of their work.
Emphasising Elements with Parentheses
Most Stable Diffusion-based models support emphasis syntax: wrapping a word or phrase in parentheses increases its weight, double parentheses increases it further. (golden eyes:1.3) means "weight this term 1.3 times more than default". This is useful when a specific detail keeps getting ignored or when you want one element to dominate.
Chaining Concepts with Commas vs. "and"
The model handles comma-separated lists of terms differently from sentences using "and". For most AI models, comma-separated terms each get equal weight. Writing "beautiful landscape, mountains, lake, fog, sunset" gives each term similar influence, whereas "a beautiful mountain landscape with a fog-covered lake at sunset" narratively connects the ideas. Experiment with both structures for different effects.
Controlling Aspect Ratio Through Composition
Even when you can't directly control aspect ratio, composition keywords influence how the model fills the frame. "Close-up portrait" tends toward square or portrait ratios. "Wide establishing shot" pulls toward landscape orientation. "Vertical composition" explicitly signals height over width. These cues help even on platforms with fixed output dimensions.
Troubleshooting Common Prompt Problems
| Problem | Likely Cause | Fix |
|---|---|---|
| Result looks generic | No style specification | Add a specific medium, art style, or photography type |
| Subject is distorted | Vague subject description | Be more specific about physical attributes and pose |
| Background overwhelms subject | No composition guidance | Add "portrait", "close-up", or "subject in foreground" |
| Mood is wrong | No lighting/atmosphere specification | Add specific lighting and colour temperature |
| Style is inconsistent across generations | Style anchor is weak | Name a specific art movement, artist, or era |
Frequently Asked Questions About Prompting
How long should my prompt be?
Aim for 20–70 words as a starting point. This is long enough to specify subject, style, lighting, and composition without diluting the weight of any single element. Very short prompts (under 10 words) leave too much to chance. Very long prompts (over 150 words) can cause later terms to be underweighted.
Should I use complete sentences or keyword lists?
Both work, and different models respond differently. Keyword lists separated by commas tend to work well for most Stable Diffusion-based models because each term gets roughly equal weight. Complete sentences work well for models with stronger natural language understanding (like DALL-E 3). Try both and notice which produces more predictable results on the platform you use.
Why does adding "masterpiece" or "award-winning" help?
These quality boosting terms work because the model was trained on images paired with text that often included such descriptions. High-quality, frequently shared images were more likely to be described with superlative language. So "masterpiece, highly detailed" statistically shifts generation toward the high-quality end of the model's learned distribution. It's a correlation, not a instruction — but the effect is real.
What is the most common mistake in prompt writing?
Forgetting to specify a style or medium. A prompt describing only a subject ("a mountain at sunset") produces generic, average results because the model defaults to its most common training output. Adding a style ("oil painting", "cinematic photography", "watercolour illustration") immediately focuses the generation on a specific aesthetic range and usually produces dramatically better results.
Can I use prompts I find online?
Yes, and this is an excellent way to learn. Sites like PromptHero.com and Civitai.com publish prompts with their outputs. Using these as starting points and modifying them to your needs is a legitimate and effective learning method — much like a chef learning from existing recipes before developing their own.
Prompt Templates by Subject Category
Here are complete, ready-to-adapt prompt templates covering the most commonly requested subject categories:
- Portraits: "[subject description], soft studio lighting, 3/4 view, shallow depth of field, professional portrait photography, 85mm lens, warm and natural tones, highly detailed, photorealistic"
- Landscapes: "[location/type], golden hour light, atmospheric perspective, layered depth — foreground detail, mid-ground subject, receding background, cinematic photography, highly detailed, dramatic sky"
- Product: "[product description], [surface — marble/wood/concrete], clean studio lighting, soft shadows, product photography, commercial, white background, sharp focus, professional"
- Conceptual illustration: "[abstract concept] depicted as [visual metaphor], [art style], [colour palette], editorial illustration, professional, detailed, [mood descriptor]"
- Architectural: "[building type], [architectural style], [time of day] lighting, sharp architectural photography, clean lines, professional, [weather/atmosphere]"
- Food: "[dish name], [hero shot angle — overhead or 45°], [surface material], [garnish description], food photography, styled, warm natural light, shallow depth of field, appetising"
Each template is a starting point — substitute the bracketed elements with your specific subject and refine from there. The structural logic (subject → context → lighting → style → quality) applies across all of them.