The order of prompts and keywords matters when generating images with Stable Diffusion. As an open-source AI model, Stable Diffusion relies on carefully crafted text prompts to produce images. Mastering prompt order and structure is key to guiding the model properly.
In this article, we’ll cover the basics of structuring Stable Diffusion prompts for better results. I’ll share my personal tips and tricks as an avid SD user.
Why Prompt Order Matters
Stable Diffusion processes prompts in a linear fashion, analyzing the text from start to finish. Putting the most important information first focuses the model’s attention where you want it. Descriptions toward the end of a long prompt tend to get less emphasis.
For example, if you specify a color at the start of your prompt, it will likely dominate the final image. But place that color descriptor after a detailed character description, and it may get overlooked entirely.
Through trial and error, you can learn how to balance prompt components to weight the elements you care about most. But it takes some experimentation to find that sweet spot!
Structuring Your Prompts
When writing Stable Diffusion prompts, it helps to break things down into logical sections focused on key elements like the subject, style, composition, and details.
Subject
Start by introducing the main topic or character. Get very specific here about exactly what you want to generate.
Good example: “A portrait of a girl with red hair, green eyes, and fairy wings.”
Bad example: “A picture of a girl.”
Style
Next, define the overall style. Reference specific artists, genres, art movements, etc. This cues the model on what “look” you’re going for.
Example: “digital art by Greg Rutkowski and Alphonse Mucha.”
Composition
Describe the pose, framing, angle, lighting, etc. Setting the scene here tells SD how to construct your image.
Example: “Looking to the side, close up shot, soft lighting.”
Details
Finally, add any small flourishes, accessories, colors or context that matter. The model will latch onto these descriptors so place them intentionally.
Example: “Surrounded by flowers, wearing a flower crown.”
Advanced Prompting Techniques
Once you get the basics down, there are a few handy tricks to further refine your Stable Diffusion prompts.
Attention Modifiers
Wrap important keywords in parenthesis or add a colon plus numerical weight to make parts of your prompt stand out.
Example: “A photo of a (fluffy:1.2) cat in a wizard hat.”
Negative Prompts
Add phrases that start with “-” to exclude unwanted elements from being generated.
Example: “-extra fingers, -mutation, -deformed”
Creative Misspellings
Introduce variation by misspelling some terms, which SD will interpret uniquely.
Example: “A scenery digital painting of a mystical forrest with unicorns.”
Prompt Structure Examples
Let’s look at a few full prompt examples showcasing proper order and formatting:
Character Portrait
“A beautiful digital painting portrait of a young woman with long blonde hair, bright blue eyes, rosy cheeks, and light freckles, wearing a flower crown and white dress, soft lighting, depth of field, by Alphonse Mucha and Greg Rutkowski.”
Fantasy Landscape
“A mystical landscape digital painting of a lush green forrest with a winding dirt path leading to a quaint village next to a crystal lake at sunset, trending on Artstation, by Greg Rutkowski, mystical lighting, highly detailed, 8k resolution.”
Still Life Scene
“An oil painting still life of a collection of musical instruments on a wooden table by a sunny window, soft natural lighting, high realism, by Rembrandt, classical Baroque style, elegant composition.”
Helpful Resources
Here are some useful sites to reference as you continue honing your Stable Diffusion prompt skills:
Stable Diffusion Prompt Guide – Prompt examples and formatting tips
LAION 400M Image Dataset – Browse examples SD is trained on
Nightcafe Studio – Creative prompt ideas
SD Discord – Connect with the SD community
Conclusion
With the right approach, Stable Diffusion can produce jaw-dropping images from text prompts alone. Carefully structuring those prompts guides the AI model to focus on the key elements you define. Start with the subject, style, composition and details in that order. Use attention modifiers, negative prompts and creative misspellings to refine your results.
I hope these Stable Diffusion prompting tips help you unlock your creativity and take your AI artistry to the next level! Let me know if you have any other questions.