AI Image Prompt Engineering for Beginners: A Complete Guide to Writing Quality Prompts

Prompts are the core input for AI image generation. Writing good prompts can multiply the quality of outputs. This article systematically explains the principles and techniques of AI image prompt engineering.

1. Basic Prompt Structure

1.1 Universal Formula

A high-quality prompt typically includes these elements:

[Subject] + [Action/Pose] + [Environment/Background] + [Style/Medium] + [Lighting/Color] + [Composition/Angle] + [Quality/Details]

Example:

A white cat sitting on a windowsill, sunlight streaming in, photorealistic style, warm soft tones, close-up shot, 8K ultra HD, depth of field

1.2 Platform Differences

Platform Language Preference Length Style Word Weight
Midjourney English best 60-200 chars High
DALL-E 3 Natural language 40-120 chars Medium
Stable Diffusion English + tagging 30-100 chars Very High

2. Advanced Midjourney Prompts

2.1 Using Parameters

Midjourney supports rich parameter controls:

/imagine prompt: cyberpunk city night --ar 16:9 --v 6 --s 750 --stylize 500
  • --ar: Aspect ratio (16:9, 4:3, 1:1)
  • --v: Model version (v6 latest)
  • --s: Stylization (0-1000)
  • --stylize: Artistic quality (0-1000)
  • --iw: Image weight (0-2, for image-to-image)

2.2 Style Reference

Use --sref to reference style images:

/imagine prompt: landscape painting --sref [style_image_URL]

2.3 Common Style Keywords

Style English Keywords
Photorealistic photorealistic, hyperrealistic, 8K, ultra detailed
Cyberpunk cyberpunk, neon lights, futuristic city
Watercolor watercolor, soft brushstrokes, paper texture
Oil Painting oil painting, impasto, canvas texture
Anime anime style, cel shading, Studio Ghibli
Pixel Art pixel art, 8-bit, retro game
3D Render 3D render, octane render, C4D

3. Stable Diffusion Prompt Techniques

3.1 Positive and Negative Prompts

Positive: (masterpiece, best quality, ultra-detailed)
Negative: (nsfw, lowres, bad anatomy, bad hands, text, error, missing fingers)

3.2 Weight Control

(red hair:1.3)  — Increase "red hair" weight
(blue eyes:0.8) — Decrease "blue eyes" weight
[tree]          — Reduce weight to 0.5x

3.3 Embeddings and LoRA

Using pre-trained Embeddings and LoRA models can significantly improve the generation of specific styles or characters.

4. DALL-E 3 Prompt Techniques

DALL-E 3 has strong natural language understanding. Suggestions:

  • Use complete sentences, not tag stacks
  • Specify elements to exclude
  • Specify colors, materials, lighting details

5. Common Issues and Solutions

Issue Solution
Distorted hands Add negative prompt: bad hands, or detail in description
Inconsistent style Reduce style keywords, focus on 1-2 core styles
Poor composition Specify angle (top-down, low-angle, close-up)
Color deviation Specify dominant colors and palette

6. Recommended Resources

  • PromptBase: Buy and sell prompts
  • Midjourney Prompt Helper: Online prompt builder
  • Civitai: Stable Diffusion model and prompt sharing community