Stable Diffusion is an AI system that generates images from text prompts. Getting good results requires carefully crafting prompts that provide enough detail and context for the AI. This allows you to guide the output while still giving the model space for creativity.
Crafting effective prompts is both an art and a science. It requires an understanding of the AI model’s capabilities and limitations. With practice, you can learn to reliably produce impressive images tailored to your creative vision.
This article provides an overview of key prompt writing principles and includes numerous examples to aid your prompt engineering endeavors.
What Makes a Good Prompt?
A good prompt clearly describes the desired image, providing key details about:
- Subject matter: The main focus like a person, animal, object etc.
- Style/medium: Painting, drawing, 3D render etc.
- Attributes: Visual qualities like color, texture, lighting etc.
- Composition: Pose, framing, background etc.
Prompts should be concise yet descriptive. Using too many words can confuse the AI, resulting in muddled outputs. On the flip side, too little detail leaves much open to interpretation.
Consistency in vocabulary also helps. Referring to the same element with different terms (e.g. woman, lady, girl) makes it harder for the AI to latch onto the key prompt aspects.
Effective prompts tend to follow a similar overall structure:
[Subject] [Style/Medium modifiers] [Attributes] [Composition]
Let’s break this down element by element, with examples for each.
Crafting Stable Diffusion Prompts
Defining Subject Matter
Clearly state the main focus of the image. For example:
- A tall woman
- An astronaut repairing a spaceship
- A still life painting of fruit
Subject descriptors help narrow down the image content. Without them results tend to be more abstract.
Specifying Style and Medium
Indicate the desired stylistic qualities or medium like:
- Oil painting
- Pen and ink drawing
- Vector art
- Low poly 3D render
You can reference artistic movements too:
- Cubist style
- Impressionist landscape
Add visual details regarding colors, lighting, textures and more:
- Wearing a red dress
- Golden hour lighting
- Rough unfinished texture
Attributes bring extra realism and control over the look of generated images.
Guide pose, framing, background and other compositional elements with phrases like:
- Portrait close up
- Sitting on a park bench
- Looking at the viewer
Composition prompts place the subject matter into a scene. This adds narrative and contextualizes the final image.
Using Negative Prompts
You can also specify things to avoid with negative prompts:
- Low quality
- Extra limbs
Negative prompts give more precision by restricting unwanted elements from being generated.
Prompt Examples and Variations
Below are some prompt examples for common subjects, each with a few variations to showcase how small tweaks make a difference.
Woman Portrait Prompts
A beautiful young woman with long blonde hair wearing a red dress, portrait
A beautiful young Latina woman with long curly black hair, portrait close up
Blue eyed pale woman with short blonde bob haircut and pearl earrings, facing right in a bust portrait
An oil painting landscape of a green mountain forest in fall with a lake, by Claude Monet
A fantasy landscape digital painting of a lush valley with waterfalls and castles, trending on artstation
Photorealistic Icelandic volcanic mountain landscape with northern lights
An arctic fox with soft white fur sitting in the snow, detailed portrait
A cute capybara made of clay wearing a top hat and monocle
A still life oil painting of a bowl of assorted tropical fruits by Henri Matisse
There are endless variations for any imaginable prompt topic. Start simple and add one descriptor at a time to hone in on your desired image.
With practice, you’ll get better at crafting prompts that produce impressive, creative images tailored to your artistic vision. Use these resources and the examples above to level up your skills. Most importantly, don’t be afraid to experiment!