AI Art Prompt Engineering Guide
Master AI art prompt writing techniques from beginner to advanced, including universal formulas, weight syntax, style keyword library and practical templates
20 min read
Created 6/1/2026
Table of Contents
Welcome to the AI Art Prompt Engineering Guide! Whether you're using Midjourney, Stable Diffusion, Seedream, or other AI art tools, mastering prompt writing is the key to getting ideal results. This tutorial covers everything from basic concepts to advanced syntax, helping you systematically improve your prompt engineering skills.
This tutorial covers:
- ✅ Basic concepts and structure of prompts
- ✅ Universal formulas and practical templates
- ✅ Advanced syntax techniques (weights, scheduling)
- ✅ Complete style keyword library
- ✅ Platform differences comparison
- ✅ Common mistakes and how to avoid them
Prompt Basics
A prompt is the bridge between you and AI art tools. Simply put, a prompt is how you describe "what kind of image you want" to the AI using text. A good prompt helps the AI understand your intentions accurately and generate results that meet your expectations.
Basic Structure of Prompts
Regardless of which AI art tool you use, prompts typically contain the following components:
- Subject: The main content of the image, such as people, animals, buildings
- Scene: The environment where the subject is located, such as forest, city, space
- Style: Artistic expression method, such as oil painting, watercolor, cyberpunk
- Quality: Image quality requirements, such as HD, 8K, fine details
- Parameters: Technical parameters such as aspect ratio, seed value
Universal Formula
Mastering a universal prompt formula can help you quickly build effective prompts. Here is a proven universal formula:
Universal Formula = Quality Tags + Subject Description + Scene/Background + Art Style + Lighting/Color + Technical Parameters
Formula Breakdown
| Component | Function | Example |
|---|---|---|
| Quality Tags | Set output quality baseline | masterpiece, best quality, ultra-detailed, 8k |
| Subject Description | Define core image content | A young woman in Hanfu holding an oil-paper umbrella |
| Scene/Background | Set environmental atmosphere | Jiangnan water town, small bridges, flowing water, misty rain |
| Art Style | Determine visual expression | Ink painting style, Gongbi, ancient Chinese aesthetic |
| Lighting/Color | Control lighting effects | Soft side lighting, warm tones, sunset lighting |
| Technical Parameters | Adjust output specifications | --ar 16:9 --v 6 |
Subject Description
The subject is the core of the image. The more specific your description, the more the AI-generated result will match your expectations.
Description Technique Comparison
| ❌ Vague Description | ✅ Specific Description |
|---|---|
| A girl | 16-year-old girl, waist-length silver hair, amber eyes, smiling expression, wearing a light blue dress |
| A cat | A fluffy orange tabby cat with green eyes, lazily lying on a windowsill basking in the sun |
| A building | Gothic cathedral, spired towers, stained glass windows, illuminated at night, magnificent and grand |
💡 Tip: Use adjective + noun combinations, adding details like quantity, color, action, and expression. Use the template: "What kind of [subject] + doing what + in what state"
Style Definition
Style determines the artistic expression of the image. AI art tools support almost all known art styles; the key is knowing how to describe them.
Common Style Types
| Style Category | Keyword Examples |
|---|---|
| Traditional Painting | Oil painting, watercolor, sketch, printmaking, Gongbi, ink wash, mural |
| Modern Art | Pop art, minimalism, abstract expressionism, surrealism, cubism |
| Digital Style | 3D render, pixel art, vector illustration, cyberpunk, vaporwave, low poly |
| Photography Style | Portrait photography, landscape photography, macro photography, long exposure, HDR, black and white |
| Anime Style | Studio Ghibli style, Makoto Shinkai style, cel shading, chibi, American comic style, pixel art |
Reference Artists/Works
You can directly reference artist names to mimic their styles:
- Classical: Van Gogh, Monet, Da Vinci, Picasso, Michelangelo
- Modern: Yayoi Kusama, Takashi Murakami, Andy Warhol
- Anime: Hayao Miyazaki, Makoto Shinkai, Yoneyama Mai, Naoko Takeuchi
- Concept Design: Sparth, Craig Mullins, Ross Tran
Technical Parameters
Technical parameters control the technical specifications of the output. Different platforms have different parameter syntax.
Universal Quality Tags
masterpiece, best quality, ultra-detailed, 8k resolution, sharp focus,
professional lighting, intricate details, highres, extremely detailed
Negative Prompts
Negative prompts are used to exclude elements you don't want:
lowres, bad anatomy, bad hands, text, error, missing fingers,
extra digit, fewer digits, cropped, worst quality, low quality,
normal quality, jpeg artifacts, signature, watermark, username,
blurry, bad feet, mutation, deformed, extra limbs, extra arms,
extra legs, malformed limbs, fused fingers, too many fingers,
long neck, cross-eyed, mutated hands, polar lowres, bad face
Advanced Syntax
Mastering advanced syntax allows you to more precisely control the AI generation process and achieve more complex effects.
Weight Control
Weight control lets you adjust the importance of various elements in your prompt.
Stable Diffusion / WebUI Syntax
| Syntax | Effect | Example |
|---|---|---|
| (word) | Weight ×1.1 | (red hair) - Enhance red hair effect |
| ((word)) | Weight ×1.21 | ((masterpiece)) - Significantly boost quality |
| (word:1.5) | Custom weight | (blue eyes:1.3) - Increase weight by 30% |
| [word] | Weight ×0.9 | [watermark] - Reduce watermark probability |
Midjourney Multi-Prompt
Use double colons to separate and assign weights:
surreal landscape::2 floating islands::1 balloons::2
# Explanation:
# surreal landscape weight is 2
# floating islands weight is 1
# balloons weight is 2
Scheduled Prompting
Advanced feature to control using different prompts at different sampling stages.
Syntax Format
| Syntax | Description | Example |
|---|---|---|
| [from:to:when] | Switch from from to to | [flowers:trees:0.7] - Change from flowers to trees after 70% |
| [to:when] | Start drawing to at when | [flowers:0.7] - Start drawing flowers after 70% |
| [from::when] | Stop from at when | [flowers::0.7] - Stop drawing flowers after 70% |
Other Advanced Syntax
| Syntax | Function |
|---|---|
| A BREAK B | Break connection between prompts, reduce contamination |
| A AND B | Blend multiple elements (e.g., two-tone hair) |
| [A|B] | Alternate drawing, generate fantastical mixed effects |
| word1_word2 | Underscore connection, prevent tokenization ambiguity |
Style Keyword Library
Here is a complete style keyword reference for quickly finding the descriptors you need.
Lighting Effects
| Effect Type | Keywords |
|---|---|
| Natural Light | golden hour, blue hour, soft diffused light |
| Artificial Light | studio lighting, neon lights, candlelight |
| Special Effects | volumetric lighting, rim light, god rays, lens flare |
| Atmosphere | cinematic lighting, dramatic lighting, moody lighting |
Perspective and Composition
| Perspective Type | Keywords |
|---|---|
| Camera Angles | portrait, full body, from above, from below, bird's eye view |
| Lens Types | wide angle, fisheye, macro, telephoto |
| Depth of Field | shallow depth of field, bokeh, deep focus |
| Composition | rule of thirds, centered, symmetrical, dynamic angle |
Materials and Textures
| Material Type | Keywords |
|---|---|
| Metal | metallic, gold, silver, brushed metal, rusty |
| Natural Materials | wooden, stone, marble, crystal, gemstone |
| Fabric | silk, velvet, lace, leather, denim |
| Special Effects | translucent, iridescent, holographic, glowing |
Mood and Atmosphere
| Mood Type | Keywords |
|---|---|
| Positive | joyful, peaceful, romantic, dreamy, whimsical |
| Negative | melancholic, mysterious, ominous, haunting |
| Intense | epic, dramatic, intense, powerful |
| Calm | serene, tranquil, meditative, zen |
Practical Templates
Here are proven practical prompt templates that you can use directly or modify according to your needs.
Template 1: Portrait
masterpiece, best quality, portrait of a [age] [gender] with [hair style] hair,
[eye color] eyes, wearing [clothing description], [expression] expression,
[scene description], [lighting description], [photography style],
[camera parameters], ultra-detailed, 8k --ar 3:4
Example:
masterpiece, best quality, portrait of a 25-year-old woman with long wavy
auburn hair, emerald green eyes, wearing an elegant Victorian dress, gentle
smile expression, sitting in a sunlit garden, soft natural lighting,
professional portrait photography, 85mm lens, shallow depth of field,
ultra-detailed, 8k --ar 3:4
Template 2: Landscape
masterpiece, best quality, [time] landscape of [location], [weather/season],
[main elements], [lighting description], [art style],
[color description], atmospheric, ultra-detailed, 8k --ar 16:9
Example:
masterpiece, best quality, sunset landscape of Japanese countryside,
autumn season, red maple trees along a winding river, golden hour lighting,
Studio Ghibli style, warm orange and red tones, atmospheric,
ultra-detailed, 8k --ar 16:9
Template 3: Concept Art
masterpiece, best quality, concept art of [subject], [style description],
[material description], [lighting description], [atmosphere description],
artstation, unreal engine 5, octane render, ultra-detailed, 8k --ar 16:9
Example:
masterpiece, best quality, concept art of a futuristic cyberpunk city,
neon-lit streets with holographic advertisements, metallic and glass surfaces,
dramatic volumetric lighting, rainy night atmosphere, artstation,
unreal engine 5, octane render, ultra-detailed, 8k --ar 16:9
Template 4: Product Photography
professional product photography of [product], [material], [background],
[lighting], [camera angle], commercial photography style,
sharp focus, ultra-detailed, 8k --ar 1:1
Template 5: Anime Illustration
masterpiece, best quality, anime style illustration of [subject],
[character description], [scene], [mood],
vibrant colors, cel shading, detailed background,
art by [artist name], ultra-detailed, 8k --ar 2:3
Platform Differences
Different AI art tools have different prompt syntax and characteristics. Understanding these differences can help you use each platform better.
Midjourney
| Feature | Description |
|---|---|
| Natural Language | Good understanding of natural language, can use sentences |
| Parameter Syntax | --ar, --v, --s, --c, --seed, etc. |
| Multi-Prompt | Use :: to separate and assign weights |
| Stylization | --stylize controls artistic level (0-1000) |
| Negative Prompt | Use --no parameter, e.g., --no text, watermark |
Stable Diffusion (WebUI/ComfyUI)
| Feature | Description |
|---|---|
| Weight Syntax | () enhance, [] reduce, (word:1.5) precise control |
| Scheduled Prompting | [from:to:when] syntax support |
| Embedding | Can use trained embedding models |
| LoRA | Support loading LoRA models to adjust style |
| Negative Prompt | Dedicated negative prompt input box |
Seedream (Jimeng AI)
| Feature | Description |
|---|---|
| Chinese Understanding | Excellent understanding of Chinese prompts, Chinese recommended |
| Smart Optimization | Built-in prompt optimization feature |
| Reference Image | Support style reference, composition reference, character reference |
| Model Versions | 3.0/4.0/4.5, newer versions have better quality |
| Negative Prompt | Support negative prompt input |
DALL-E 3
| Feature | Description |
|---|---|
| Natural Language | Best for natural language long descriptions |
| Text Rendering | Strongest text rendering capability |
| Parameters | Fewer parameter supports, mainly rely on description |
Common Pitfalls
Here are common mistakes beginners make and their solutions.
Common Mistakes
| ❌ Wrong Approach | ✅ Correct Approach |
|---|---|
| Longer prompts are better | Clear structure matters more than length, usually 50-100 words are enough |
| Using natural language long sentences | Use comma-separated keyword lists, avoid complex sentences |
| "Many flowers", "Big scene" | "Cherry blossoms covering 50% of area", "Subject 30%, background 70%" |
| Ignoring negative prompts | Must add low quality, deformed fingers, watermark exclusions |
| Token exceeds 75 | Control single prompt ≤75 tokens (about 50-60 words) |
| Contradictory keywords | Avoid contradictory descriptions like "daytime" and "night" simultaneously |
Efficiency Tips
- Use AI to help write prompts: Send target images to ChatGPT/DeepSeek and let it generate prompts
- Build personal keyword library: Collect commonly used quality and style word combinations
- Iterate and optimize: Don't expect perfect results in one generation, adjust prompts based on output
- Record seed values: Record seed when you get good results for reproducibility
- Learn from excellent works: Analyze prompt structures of excellent works in the community
Problem Solving Quick Reference
| Problem | Solution |
|---|---|
| Deformed fingers | Add "bad anatomy, bad hands, extra fingers" to negative prompts, or use "closed hands" to avoid drawing hands |
| Distorted face | Add "perfect face, detailed face", or use face restoration plugins |
| Gibberish text | Add "text, watermark, signature" to negative prompts, or use dedicated text rendering tools |
| Inconsistent style | Reduce conflicting style words, use weights to control main style |
| Too abstract image | Add specific descriptive words, reduce --stylize value, add "realistic" |
🎉 Congratulations on completing this tutorial!
You now have the core skills of AI art prompt engineering. Remember, prompt writing is a practical skill that requires practice to truly master. Suggestions:
- Start by imitating prompts from excellent works
- Change only one variable at a time and observe effect changes
- Build your own prompt template library
- Follow the Promptbox community to learn from other creators
- Share your experiences and discoveries