Stable Diffusion is a powerful AI image generation model that creates images from text prompts. Carefully crafted prompts are key to generating high-quality and consistent images in Stable Diffusion. This article provides guidance on the most effective prompt keywords to use.
Defining the Subject
The first step in writing an effective prompt is to clearly define the subject of the image. Use specific nouns and adjectives to describe exactly what you want the AI to generate.
Here are some examples of defining a subject:
- A photo of a calico cat with green eyes
- A realistic digital painting of a dragon in flight over a fantasy landscape
- A 3D render of a futuristic city with flying cars and holograms
Describing Features
After defining the subject, describe its key features in more detail using adjectives and descriptive phrases:
- A calico cat with fluffy fur, sharp claws, and alert, perked-up ears
- A realistic digital painting of a fiery red dragon with enormous, bat-like wings in flight over lush green hills and snow-capped mountains
- A 3D render of a vibrant, bustling futuristic city with sleek, oval-shaped flying cars zooming by tall, glass-covered skyscrapers displaying bright, neon holograms
Setting a Style
Next, set the artistic style using names of art movements or specific artists:
- A photo of a calico cat in the style of impressionist painting
- A realistic digital painting of a dragon in the style of fantasy book cover art by John Howe
- A 3D render of a futuristic city in the style of 80s retro-futurism and cyberpunk
You can also reference more than one style:
- A 3D render in the style of art deco architecture combined with cyberpunk neon lighting
Adding Quality Descriptors
Include descriptors that set quality expectations:
- Extremely detailed
- Intricately painted
- Cinematic lighting
- Vibrant colors
- Sharp focus
- High resolution
- 8k render
Example prompt:
An intricately painted, extremely detailed digital portrait of an elf warrior with long blonde hair, pointy ears, and glowing green eyes, wearing silver armor. Rendered in sharp focus and vibrant colors at 8k resolution in the style of fantasy art by Greg Rutkowski and Simon Stålenhag. Guiding Composition
Guide the AI’s composition with instructions on:
Camera Angle
- Straight-on portrait
- Profile portrait
- Bird’s-eye view
- Worm’s-eye view
Framing
- Extreme close-up shot
- Medium close-up shot
- Full body shot
- Wide landscape shot
Orientation
- Vertical portrait orientation
- Horizontal landscape orientation
Example prompt with composition guidance:
A full body shot of an astronaut standing on the surface of Mars, horizontal composition, extremely detailed and vibrantly colored digital painting by Greg Rutkowski with a cinematic atmosphere and lighting. The astronaut looks up in awe at the starry skies, Earth visible in the distance. Rendered at 8k resolution.Negative Prompts
Add “negative prompts” to specify what you don’t want to see in the image:
- ugly
- poorly drawn
- bad anatomy
- extra fingers
- mutation
- mutilation
- disfigured
Negative prompts help narrow down the image search space so you have more control over the output.
Example with negative prompts:
An intricately painted portrait of an elf warrior with flowing hair, rendered in oil paints in the style of Greg Rutkowski. Extremely detailed and vibrant colors, no mutations or ugly distortions.Test and Refine
Be prepared to test and refine your prompts through multiple iterations to achieve your desired result. Adjust descriptors, add more details, tweak the style references, and add negative prompts until the output matches what you envisioned.
Prompt engineering is an iterative process but the above components form the core building blocks for Stable Diffusion prompts.
Prompt Keyword Categories
Here is a summary of the key prompt keyword categories:
Subject Definition Keywords
- Nouns – define the topic
- Adjectives – describe features
Style Keywords
- Art movements
- Artist names
- Genres
Quality Keywords
- Lighting
- Level of detail
- Resolution
- Focus
- Color
Composition Keywords
- Camera angle
- Framing
- Orientation
Guiding Keywords
- Negative prompts – exclusions
Prompt Engineering Resources
Here are some useful resources for learning more about crafting better Stable Diffusion prompts:
Stable Diffusion Prompt Engineering Guide
Best Practices for Stable Diffusion Prompts
Common Prompt Engineering Mistakes
Stable Diffusion Prompt Anatomy
Conclusion
Carefully engineered prompts are crucial to generating consistent, high-quality AI images with Stable Diffusion. Defining the subject, describing features, setting artistic style, adding quality descriptors, guiding composition, and using negative prompts gives you control over the output. Prompt engineering takes practice but by leveraging the right keywords you can produce amazing AI art.