Prompt Techniques That Work 💡

[admin]

Here's the collected wisdom on what prompt techniques actually deliver results across different models and styles.

The fundamental structure:

Subject + Style + Lighting + Composition + Quality

This formula works across all models. "Portrait" is weak. "Close-up portrait, dramatic Rembrandt lighting, oil painting style, rich colors, masterpiece quality, 8K" is strong.

Photography style prompts:

Camera and lens specs control everything. Compare these:

"Shot on iPhone" – casual, wide angle, everything in focus "Shot on Hasselblad 500C, 80mm, f/2.8" – medium format look, beautiful depth "Shot on Canon 5D, 85mm, f/1.4" – portrait lens, creamy bokeh "35mm Kodak Portra 400 film" – warm tones, natural grain

These terms work even for non-photo subjects. Try "fantasy landscape, shot on medium format film" – creates specific aesthetic.

Lighting terminology:

"Golden hour" – warm, soft, flattering "Blue hour" – cool, moody, twilight "Rembrandt lighting" – dramatic triangle under eye "Rim lighting" – backlit edge highlight "Soft diffused light" – even, flattering, professional "Harsh overhead lighting" – unflattering, dramatic shadows

Study cinematography lighting to learn terms AI responds to.

Artist style references:

"Greg Rutkowski" – epic fantasy digital painting "James Gurney" – realistic painted illustration "Alphonse Mucha" – Art Nouveau decorative "Simon Stålenhag" – melancholic sci-fi realism "Hayao Miyazaki" – Studio Ghibli animation style

Browse ArtStation to discover artists whose aesthetic you want. Reference them in prompts for instant style direction.

Quality markers:

These boost output quality:

"8K, highly detailed, sharp focus" – technical quality "Trending on ArtStation" – pushes toward polished digital art "Award-winning, masterpiece" – quality indicators "Professional photography" – commercial standard "Intricate detail, hyperrealistic" – maximum detail

Composition terms:

"Close-up" – fills frame "Medium shot" – waist up "Wide shot, establishing" – shows environment "Dutch angle" – tilted, dynamic "Rule of thirds composition" – balanced framing "Symmetrical composition" – centered, formal

Style-specific prompts:

Cyberpunk: "neon lighting, rain-slicked streets, blade runner aesthetic, purple and cyan color scheme, dystopian, night scene"

Fantasy: "epic fantasy, dramatic lighting, castle background, magical atmosphere, trending on ArtStation, detailed armor, heroic pose"

Product photography: "studio lighting, white background, commercial photography, highly detailed, sharp focus, professional product shot, clean composition"

Anime: "anime style, cel shading, vibrant colors, Studio Ghibli aesthetic, detailed character design, dynamic pose"

Vintage: "1970s aesthetic, faded colors, film grain, retro typography, vintage photograph, nostalgic mood"

Model-specific strategies:

Midjourney tips:

• Photography terms work great: "shot on Hasselblad"
• Use --stylize to control artistic interpretation
• Reference decades: "1980s, 1950s aesthetic"
• Combine artists: "Greg Rutkowski meets Alphonse Mucha"

DALL-E tips:

• Be conversational and explicit
• Great at text: "include the text 'Hello World' in elegant font"
• Describe scenes naturally: "A coffee shop on a rainy day, warm lighting through windows, cozy atmosphere"

Stable Diffusion tips:

• Use negative prompts: "ugly, distorted, low quality, blurry, watermark"
• Emphasis syntax: (dramatic lighting:1.3) for more weight
• Quality prompts at start: "masterpiece, best quality, highly detailed"

The bracket technique:

Organize complex prompts with clarity:

[Subject: ancient dragon] [Style: fantasy oil painting] [Lighting: dramatic side lighting with rim light] [Composition: close-up of face, shallow depth of field] [Quality: 8K, highly detailed, trending on ArtStation]

This helps you remember what each part does.

Prompt libraries:

PromptHero – search 10M+ prompts by style or subject Lexica.art – Stable Diffusion specific search OpenArt – prompt suggestions and style browse Midjourney Community Feed – see prompts for all public generations

What doesn't work:

Being vague – "make it cool" means nothing Over-complicating – 200-word prompts confuse models Contradictions – "photorealistic anime" fights itself Ignoring model strengths – asking DALL-E for artistic interpretation when Midjourney excels at it

Testing methodology:

Change one element at a time. If you modify lighting AND style AND composition simultaneously, you won't know what worked. Test systematically: generate baseline → change only lighting → compare → change only style → compare.

Collecting your prompts:

Build a personal prompt library. When something works, save it with notes. Organize by: style (cyberpunk, fantasy, product), subject (portraits, landscapes, objects), model (Midjourney, DALL-E, SD).

Use a note-taking app or spreadsheet. Future you will thank present you.

Share prompts that worked here. Post before/after comparisons showing what changed. Ask why a prompt isn't delivering expected results. When you discover a technique, document it for others.