Stable Diffusion Prompt Learning

Stable diffusion is a powerful AI system for generating images from text descriptions. Crafting effective prompts is key to getting good results from stable diffusion. This article provides examples of stable diffusion prompts across different categories to help guide prompt engineering efforts.

Mastering stable diffusion prompts enables creating imaginative, high-quality AI art. Prompt engineering is an iterative process of refining prompts through trial and error to produce better images. Let’s explore prompt examples you can build upon.

Scenes

Scenes depict environments and arrangements of elements. Specify details like the number of subjects, poses, colors, lighting, and style.

A tall oak tree with a tire swing, next to a lake at sunset, matte painting trending on ArtStation
A steampunk airship flying over a fantasy city landscape, by Greg Rutkowski and Thomas Kinkade
Three cats playing with balls of yarn, cute cartoon animation still frame
An astronaut floating in space with the Milky Way galaxy behind them, National Geographic photography
A robot made of old computer parts, dramatic side lighting, sharp focus, octane render

Characters

Characters center on depicting individuals with visual details. Define features, clothing, expressions, species, etc.

A strong female warrior with blue skin and glowing tattoos wielding an icy sword, digital art
A goblin wearing steampunk goggles and a top hat, holding a potion, fantasy illustration
An anthropomorphic red panda wearing a flower crown, soft lighting, cute Disney style
A wise old wizard with a long white beard wearing purple robes, dramatic portrait
A female astronaut smiling proudly in front of an American flag, photorealistic

Styles

Emulate artistic styles by referring to genres, artists, movements, and qualities. Mix multiple for hybrid styles.

A fruit still life by Caravaggio, baroque realism, chiaroscuro
The Eiffel Tower in an abstract cubist style with bright colors like Juan Gris
A landscape of Machu Picchu, Andean folk art by Joaquín López Antay
A robot walking a dog, Pixar animation concept art by Ralph McQuarrie
A deer in a forest, impressionist oil painting with thick visible brush strokes 

Editing

Guide edits by describing differences from a source image. This enables iterative improvements.

The same portrait but with short dark hair instead of long blonde hair
The same landscape photo but make it a beautiful autumn scene with red and yellow trees  
The same machine but with more gears, pipes, and smoke effects added
The same alien character but with more realistic slimy skin and big bright yellow eyes
The same fruit basket image but viewed from above at a Dutch angle

Mashups

Combine disparate concepts and cross over genres for creative fusions. This tests imagination.

A steampunk robot cowboy holding a laser pistol riding a horse in space
A cute baby dragon wearing a flower crown holding a pink heart shaped box of chocolates 
The Mona Lisa as an anime character with pink hair and blue eyes  
A luxury sports car that is also a crab with pincers for wheels and a shell for the chassis
The starship Enterprise inside an Art Deco style building docked at a space station

Text Insertion

Insert custom text dynamically into images like logos, captions, dialogue, codes, etc.

A phone app icon that says "Food Finder" in a bold serif font 
A motivational poster with the quote "Attitude is a little thing that makes a big difference"  
A comic strip of two cats talking, one says "Did you eat all the treats again?"
A computer code snippet that says "print('Hello World')" with syntax highlighting  
A technical diagram labeled with "Neural Network Architecture" in a futuristic style

Prompt Structure

Well-structured prompts improve coherence and reduce ambiguity. Organize keywords logically.

Subject terms: Define the main topic first
Style terms: Then describe artistic styles
Modifier terms: Next add key details like poses, colors, lighting etc
Relationship terms: Finally specify relationships between elements

Prompt Length

Stable diffusion can process prompts with ~200-300 tokens. More tokens allow more details at the cost of creativity. Find the ideal balance.

75-150 tokens – Very short prompts have more randomness
150-200 tokens – Balances constraints with flexibility
200-300 tokens – Very detailed prompts leave little to imagination

Prompt Sources

Browse these sites to discover prompts made by other users for inspiration:

Analyze prompts by removing parts to understand their impact. Build your own style by remixing elements from prompts you like.

Conclusion

I hope these stable diffusion prompt examples across common categories provide some ideas to incorporate into your own prompts. Prompt engineering is an iterative process, so keep practicing and learning from results. Share your findings with the community to collectively expand what’s possible.