AI Image Generation Prompts for Designers: The Ultimate Cheat Sheet

The difference between a mediocre AI image and a stunning one almost always comes down to the prompt. Prompt engineering for image generation is a skill that combines technical knowledge of how diffusion models interpret language with artistic vocabulary and compositional thinking. This comprehensive cheat sheet covers every major aspect of AI image prompting across all major platforms.

Universal Prompt Structure

The most effective prompts follow this formula: [Subject] + [Style/Medium] + [Composition] + [Lighting] + [Color] + [Mood] + [Technical Parameters]. Each element builds on the previous one to create increasingly specific and controlled outputs.

Subject Description

Be specific and descriptive. Instead of “a dog,” write “a golden retriever puppy sitting in autumn leaves, looking directly at camera, tongue out.” Include: species/type, action/pose, expression, surrounding objects, and relationship to environment.

Style and Medium

This is where artistic vocabulary matters most. Common effective terms: Photography: “DSLR photo, Canon EOS R5, 85mm lens, shallow depth of field, bokeh.” Digital art: “digital painting, highly detailed, artstation trending.” Traditional art: “oil painting on canvas, visible brushstrokes, impasto.” 3D render: “Octane render, cinema 4D, volumetric lighting, subsurface scattering.” Illustration: “editorial illustration, flat design, limited color palette.”

Lighting Vocabulary

Lighting dramatically changes mood and quality. Key terms:

Direction: backlit, side-lit, rim lighting, Rembrandt lighting (classic portrait lighting with triangular cheek shadow), butterfly lighting (overhead, creates butterfly shadow under nose), split lighting (dramatic half-face illumination).

Quality: soft diffused light, harsh directional light, golden hour, blue hour, overcast, studio lighting, natural window light.

Special effects: volumetric rays (god rays), neon glow, bioluminescence, caustics (light patterns through water/glass), lens flare, light leak.

Composition Terms

Camera angles: bird’s eye view, worm’s eye view, eye level, Dutch angle (tilted), over-the-shoulder, aerial drone shot.

Shot types: extreme close-up (macro), close-up (face/detail), medium shot (waist up), full body, wide establishing shot, panoramic.

Composition rules: rule of thirds, centered composition, symmetrical, leading lines, frame within a frame, negative space.

Platform-Specific Tips

Midjourney

Midjourney responds well to artistic references and mood descriptions. Key parameters: --ar 16:9 (aspect ratio), --s 750 (stylization: 0=literal, 1000=artistic), --c 25 (chaos: variation between outputs), --q 2 (quality/detail level), --style raw (less Midjourney aesthetic, more literal), --cref [url] (character reference for consistency), --sref [url] (style reference). Midjourney Documentation

DALL-E 3

DALL-E 3 excels at following detailed compositional instructions and rendering text. It interprets natural language more literally than Midjourney. Tips: be very specific about spatial relationships (“the cat is sitting ON TOP OF the piano, NOT inside it”), specify text exactly as you want it rendered (“a sign that reads ‘OPEN 24 HOURS’ in red neon letters”), use descriptive paragraphs rather than keyword lists.

Stable Diffusion

Stable Diffusion benefits most from technical prompt syntax. Use quality boosters: “masterpiece, best quality, ultra detailed, sharp focus, 8k.” Use negative prompts to exclude unwanted elements: “ugly, deformed, blurry, low quality, watermark, text, signature.” Weight tokens with parentheses: “(sunset:1.4)” amplifies sunset’s influence. Use CLIP skip, CFG scale, and sampling steps for fine control. Stable Diffusion prompt guide

Style Modifiers Library

Photorealistic: hyperrealistic, photorealistic, ultra HD, 8K, RAW photo, DSLR, 35mm film, analog photography, Kodak Portra 400, Fujifilm.

Artistic movements: Art Nouveau, Art Deco, Bauhaus, Impressionism, Surrealism, Pop Art, Minimalism, Brutalism, Vaporwave, Cyberpunk.

Digital art: concept art, matte painting, digital illustration, 3D render, low poly, pixel art, isometric, retro game art, synthwave.

Traditional: watercolor, oil painting, charcoal drawing, ink wash, woodcut, linocut, screen print, risograph, Japanese woodblock print (ukiyo-e).

Color Palette Prompts

Warm palettes: “warm color palette, golden tones, amber, rust, terracotta.” Cool palettes: “cool blue tones, teal, slate, silver.” Moody: “dark and moody, desaturated, muted earth tones, cinematic color grading.” Vibrant: “vivid saturated colors, electric, neon accents, high contrast.” Pastel: “soft pastel colors, dreamy, muted, cotton candy palette.” Monochrome: “monochromatic, single color study, tonal variation.”

Common Prompt Mistakes

Too vague: “a beautiful landscape” gives the AI no direction. Add specific elements, time of day, weather, and style.

Contradictory terms: “photorealistic watercolor painting” confuses the model. Choose one medium.

Too many subjects: “a dog, a cat, three birds, a fish, and a horse in a field” usually produces chaotic results. Focus on 1–2 subjects.

Ignoring negative prompts: In Stable Diffusion, not specifying what you don’t want often results in common AI artifacts (extra fingers, watermarks, blurry backgrounds).

FAQ

How long should my prompt be?
Midjourney works best with 30–75 words. DALL-E 3 handles up to 400 characters well. Stable Diffusion can process very long prompts but gives more weight to words at the beginning—front-load important terms.

Do prompt keywords have order importance?
Yes. Most AI models give more weight to earlier words. Place your most important descriptors first. In Stable Diffusion, you can use explicit weighting syntax to override this.

How do I get consistent characters across multiple images?
Use character reference features (Midjourney –cref), maintain identical character descriptions across prompts, or use ControlNet/IP-Adapter in Stable Diffusion. For production work, training a custom LoRA model on your character ensures the highest consistency.