Stable Diffusion is a powerful AI image generation model that creates images from text prompts. Getting good results requires carefully crafting prompts that clearly communicate what you want the AI to generate.
This article provides useful prompt engineering techniques and examples to help you create better Stable Diffusion images.
Prompt Structure
A good Stable Diffusion prompt generally has a simple structure:
- Description of the desired image
- Artistic style or medium
- Any additional details
Here is an example prompt:
A scenic photograph of a winding path through a lush green forest in the style of impressionist oil painting
Breaking this down:
- “A scenic photograph…” describes the desired image
- “impressionist oil painting” specifies the artistic style
- Additional details like “lush green forest” and “winding path” provide extra guidance
Descriptions
The most important part of a Stable Diffusion prompt is the description of what you want to generate.
Be as specific as possible – provide relevant details about the subject, setting, number of subjects, etc. Ambiguous prompts produce random results.
Styles
Specifying an artistic style, medium or specific artist directs Stable Diffusion’s output. For example:
- oil painting
- watercolor
- pen and ink drawing
- Ukiyo-e
- in the style of Van Gogh
Guiding Details
Additional descriptive details guide Stable Diffusion’s image generation process towards your desired result. For example:
An oil painting of a red fox with lush green forest in the background
Here, details like “red fox”, “lush green forest” make the prompt more specific.
Repeat key details later in the prompt to emphasize them.
Useful Prompt Writing Techniques
Here are some key techniques for writing better Stable Diffusion prompts:
Iterative Refinement
Iteratively tweak prompts and observe results to land on something you like. Refine descriptions, vary artistic styles, add/remove details – experiment freely.
Test on Low Resolution
Test prompts on low resolution like 256×256 first before generating high res images. This saves time and credits.
Weight Key Concepts
Repeat key prompt concepts at the start and end of your prompt to weight them as more important.
A photo of a red fox in a forest, red fox resting in a forest clearing
Avoid Ambiguous Wording
Use definite words and descriptions – “a photo of a man” instead of “some people”. Definiteness reduces randomness in generated images.
Helpful Prompt Keywords
Here are some helpful keywords to use in Stable Diffusion prompts:
Subject Matter: self-portrait, person, man, woman, child, macro, close-up, full-length, cutout, isolated
Composition: centered, rule of thirds, golden ratio, symmetric, profile, silhouettes, off-center
Camera: wide angle, fish eye, tilt shift, shallow depth of field, motion blur, lens flare
Lighting: soft lighting, hard lighting, rim lighting, backlighting, chiaroscuro, cinematic
Color: monochromatic, duotone, tri-tone, vibrant, muted, pastel, earth tones, complementary colors
Styles:
- Realism: photorealistic, sharp focus, high detail, high resolution
- Abstract: geometrical, cubes, spheres
- Sketchy: line drawing, contours, pen and ink, charcoal
- Painting: watercolor, oil, acrylics, airbrush
- Graphic: vector, low-poly, pixel art
Use these keywords in your prompts to steer the AI’s image generation.
Useful Websites for Stable Diffusion
- Stable Diffusion Prompts – Search and browse top community prompts
- Automatic1111 Webui – Local UI for Stable Diffusion
- Lexica – Prompt ideation and analysis tool
- Nightcafe Creator – Online Stable Diffusion sandbox
Hopefully these prompt engineering tips and examples help you create better AI images with Stable Diffusion. Feel free to experiment endlessly and find what works for you!