Technique Guide

Image Generation Prompting

Image generation models — Midjourney, DALL-E 3, Stable Diffusion — require a different prompting approach than text models. Learning the vocabulary of style modifiers, composition terms, and model-specific syntax unlocks dramatically better results.

The Anatomy of an Image Prompt

Strong image prompts typically include: the subject, the style/medium, lighting, composition, and quality modifiers. The order matters less than the presence of each component.

Prompt formula

[subject] + [setting/context] + [style/medium] + [lighting] + [composition] + [quality]

Weak: “a person sitting in a cafe”

Strong: “a woman reading a book in a Parisian sidewalk cafe, warm afternoon light streaming through windows, oil painting style, impressionist brushwork, soft focus background, highly detailed”

Essential Modifier Categories

Style: photorealistic, oil painting, watercolor, anime, concept art, isometric illustration

Lighting: golden hour, dramatic side lighting, studio lighting, cinematic lighting, neon glow

Camera: wide angle, telephoto, macro, aerial view, fisheye, portrait lens

Mood: moody, ethereal, vibrant, dystopian, serene, cinematic

Quality: highly detailed, 8k, masterpiece, sharp focus, professional photograph

Composition: rule of thirds, centered subject, dynamic diagonal, negative space

Model-Specific Tips

DALL-E 3 (OpenAI)

• Handles natural language well — write in clear sentences, not keyword lists
• Explicitly include style: “in the style of a product photograph” or “illustrated in a flat design style”
• Use it via ChatGPT for iterative refinement — ask it to “make it more dramatic” or “change the background to...”
• Good for complex scenes with text and logos (better text rendering than Midjourney)

Midjourney

• Uses comma-separated keywords: “futuristic city, cyberpunk, rain, neon lights, cinematic”
• Append --ar 16:9 for aspect ratio, --v 6 for latest model
• Use --no [elements] for negative prompting: --no blur, text, watermark
• Reference artists for consistent style: “in the style of James Gurney”

Stable Diffusion

• Use negative prompts explicitly in the negative prompt field: “blurry, low quality, distorted, extra limbs”
• Adjust CFG scale (7–12 typical): higher = more prompt adherent, lower = more creative
• Model checkpoints matter hugely — different checkpoints excel at different styles
• Use LoRA adapters for consistent character or style across multiple images

Articles

Use Case

Prompting for Marketing

Image generation as part of a broader AI-powered marketing workflow — from ideation to output.

Foundations

What is Prompt Engineering?

The foundational principles — specificity, context, format — apply to image prompting too.

Technique

Zero-Shot vs Few-Shot Prompting

Image generation models benefit from descriptive prompts — understanding zero-shot helps.

Related Lessons

Structured lessons on multimodal prompting and related techniques.

Intermediate

Multimodal Prompting

Beginner

Formatting Output

Beginner

Giving Context

Learn Multimodal Prompting

The Intermediate track includes a full lesson on multimodal prompting — text-to-image, image understanding, and cross-modal reasoning.

Go to Multimodal Lesson