Stable Diffusion is an AI image generation model that creates images from text prompts. Writing effective prompts is key to generating high-quality images that match your creative vision. Here are some tips and examples for crafting great Stable Diffusion prompts.
Describe the Subject in Detail
Start your prompt by clearly describing the main subject of the image. The more descriptive details you provide, the better.
Good prompt: A close up portrait of a young woman with long blonde wavy hair, bright green eyes, rosy cheeks and red lips. She’s wearing a pink sweater and has a warm smile.
Bad prompt: A woman.
Set the Scene
After describing the subject, set the scene by describing the background, environment, lighting, etc. This helps establish the overall mood and aesthetic.
Example: A close up portrait of a young woman with long blonde wavy hair, bright green eyes, rosy cheeks and red lips. She’s wearing a pink sweater and has a warm smile. Soft natural lighting. Out of focus trees in background.
Specify Art Style
Naming a specific art style helps steer the image generation. Styles like realism, impressionism, art nouveau, etc.
Example: A close up portrait of a young woman with long blonde wavy hair, bright green eyes, rosy cheeks and red lips in the style of renaissance oil painting. She’s wearing a pink sweater and has a warm smile. Soft natural lighting. Out of focus trees in background.
Use Adjectives
Use descriptive adjectives to add color, texture, lighting and other qualities. Prioritize 1-3 main adjectives. Too many can confuse the AI.
Good: A majestic snow-capped mountain reflecting in a perfectly still turquoise lake at sunrise.
Bad: A tall, grand, awe-inspiring, monumental, imposing, breathtaking, colossal snow-covered mountain reflecting in a mirror-like, tranquil, peaceful, quiet, undisturbed turquoise lake at the first light of dawn.
Add Useful Keywords
Certain keywords can help steer the image generation like:
- Intensity: high/low detail, intricate, complex, fuzzy, blurry
- Viewpoint: aerial view, close up, panoramic
- Lighting: soft lighting, rim lighting, cinematic lighting
- Quality: photorealistic, hi-res, unreal engine, 8k resolution
Provide Useful Negatives
Adding negatives helps prevent unwanted elements from appearing.
Example: A photo of a cute puppy playing in the grass. No humans. No cars or buildings. Bright sunny day.
Test and Refine
Experiment with different wording and syntax to see what works best. Refine prompts over multiple generations.
Before: An oil painting of a bowl of fruit including apples, oranges and bananas.
After: A lush, vibrant still life oil painting of a ceramic bowl brimming with brightly colored apples, oranges, and bananas by Henri Matisse.
Use Samplers
Samplers like Euler a, Euler an, DPM 2 a, DPM 2 an provide different image generation algorithms. Test which looks best.
Example: A lush, vibrant still life oil painting of a ceramic bowl brimming with brightly colored apples, oranges, and bananas by Henri Matisse. Euler a.
Vary Aspect Ratio
The default 4:3 aspect ratio may not suit all images. Try 16:9, 2:3, 9:16, etc.
Example: A lush, vibrant still life oil painting of a ceramic bowl brimming with brightly colored apples, oranges, and bananas by Henri Matisse. Euler a, 2:3 aspect ratio.
Categories of Useful Prompt Keywords
Here are some common categories of keywords that are helpful for crafting Stable Diffusion prompts:
Subject Keywords
- man, woman, child, baby, portrait, selfie
- objects: car, train, airplane, food, flower, robot
- animals: dog, cat, horse, lion, eagle
Composition Keywords
- foreground, background, centered, rule of thirds, leading lines
Style Keywords
- realism, cartoon, anime, cubism
- cinematic, graphic novel, impressionism
Lighting Keywords
- soft lighting, hard lighting, rim lighting, raking light, cinematic lighting, chiaroscuro, high key, low key
Viewpoint Keywords
- aerial view, satellite view, straight on, side view, close up, wide angle
Quality Keywords
- high resolution, photorealistic, unreal engine, extremely detailed, Depth of Field (DOF)
Prompt Formulas
Rather than starting from scratch each time, you can start with prompt formulas or templates with placeholders to fill in.
Simple Portrait Formula
A ${style} portrait of a ${gender} with ${hair_color} ${hair_length} hair, ${eye_color} eyes and ${other_features}. ${subject} is wearing ${clothing} and has a ${mood} expression. ${lighting}, ${background}.
Detailed Character Formula
An extremely detailed digital painting of a ${style} ${character_name}, a ${character_description} with ${physical_traits}. ${character_name} is shown ${pose} while ${action} against a background depicting ${scene}. Trending on ArtStation.
Simple Still Life Formula
A ${style} painting of ${objects} arranged on a ${surface_material} ${surface_shape}. ${lighting}, ${background}.
Conclusion
Writing effective prompts for Stable Diffusion takes experimentation, but following the tips above should set you on the path to generating impressive, creative images. Describe subjects clearly, set the scene, specify style and quality, provide useful negatives and don’t be afraid to test variations. With practice, you’ll be crafting prompts like a pro.
Let me know if you have any other questions!