Stable Diffusion Prompt Workflow

Crafting effective prompts is key to getting the most out of AI image generators like Stable Diffusion. Prompt engineering is an art that requires experimentation, patience, and an understanding of how these models work. In this guide, I’ll walk you through my personal prompt creation workflow – from researching keywords to iteratively refining prompts for the best results.

Finding Keyword Ideas

The first step is gathering potential keywords to include in your prompt. I typically start by browsing prompt sharing sites like PromptHero and Lexica to find ideas. You can search prompts by category or popularity to see what’s working well for others.

I also look at prompt breakdowns in Reddit threads and Discord channels dedicated to Stable Diffusion. People often share the individual components that went into their successful prompts. This helps me learn new stylistic keywords and composition terms to try.

Drafting a Base Prompt

After collecting a long list of potential keywords, I draft a starting prompt that captures the basic subject, style, and details I want. I aim for 3-5 descriptive sentences as a base, making sure to leverage the keywords gathered earlier.

For example:

An oil painting portrait of an astronaut sitting on a crescent moon, trending on artstation. Dramatic lighting from below casts shadows across his spacesuit. He gazes longingly back at a tiny, bright blue Earth among the stars.

I try to use specific emotional descriptors (gazes longingly) over vague ones (looking at) since they provide more direction to the AI.

Choosing a Model

With my starting prompt, I now decide which Stable Diffusion model to use. The model architecture impacts the image quality and style significantly.

I test most prompts on a base model like Stable Diffusion 2 first. But for anime/cartoon styles, I use a model fine-tuned on those domains instead right away. This saves time rather than trying to coerce the base model into working.

Refining the Prompt

Now begins the iterative process of refining the prompt! I start by generating a batch of images to see how my prompt translates visually.

Based on the results, I tweak the wording to steer the images towards my vision. Common adjustments include:

  • Removing ambiguous descriptors
  • Introducing more creative keywords gathered earlier
  • Adding negative prompts to suppress unwanted elements (e.g. ugly, poorly drawn)
  • Specifying a different lighting style or camera angle

I repeat this process until I’m happy with the resulting images – which often takes many generations. Refining prompts is key to quality results with Stable Diffusion.

Share Your Creations!

With a polished prompt dialed in, I can now easily generate dozens of stunning images showcasing my idea!

I share my best AI-generated artworks on social media and portfolio sites. It’s exciting to let imagination run wild and see your visions come to life through prompts.

And that wraps up my personal stable diffusion prompt creation process! Remember that crafting excellent prompts requires ample experimentation and patience as you learn what makes the AI tick. But it enables you to unlock amazing creative potential.

Useful Prompt Engineering Resources: