Getting started with Stable Diffusion prompts

What are Prompts in Stable Diffusion?

Prompts are text descriptions that tell Stable Diffusion what image to generate. They act as instructions that guide the AI model to create visual content matching your vision. When you type a prompt, the system converts your words into numerical data that controls the image generation process.

How Prompts work

Stable Diffusion transforms your text into tokens through a process called tokenization. The model starts with random noise and gradually refines it over multiple steps, with each step guided by your prompt’s encoded information. This continues until a coherent image emerges that matches your description.

Prompt structure plays a crucial role in determining the final output. Word order significantly impacts the generation process, with terms placed at the beginning of prompts receiving greater emphasis from the model. Specificity in your descriptions leads to more predictable and refined results, as detailed prompts provide clearer guidance to the AI. Effective prompts typically include medium, medium boosters, subject, subject details, environment, and additional modifiers.

Ways of Prompting

Stable Diffusion accepts two distinct prompting styles: natural language descriptions and keyword-based instructions.

Natural language prompts are describing what you want in complete sentences, such as:

A gorgeous brunette woman standing on a balcony on a warm summer evening

Keyword-based function more like list of things to include in the generated image, such as:

woman, gorgeous, brunette, standing, balcony, summer, evening

Mixing is also possible:

a woman standing on a balcony, gorgeous, brunette, in a warm summer evening

In both ways you can achieve good results. Yes.

Positive and Negative Prompts

Positive prompt describe what you want to see in the generated image. Negative prompt describe what you don’t want to see. They work together.

Imagine that you want to buy a brand new car. Let’s say it is a red Honda Civic, 2-doors hatchback. Now let’s turn that vision into a prompts.

Positive prompt:

red car, honda civic, hatchback, 2 doors, brand new

Negative prompt:

4 doors, sedan, classic, american car, ugly, old, destroyed, broken, dirty, crashed

Keywords

Adjusting Emphasis

Stable Diffusion allows you to adjust the attention to each keyword. You have several options for changing emphasis. I recommend using the following format: (keyword:1.2) that increases emphasis to 1.2 or (keyword:0.5) that decreases emphasis to 0.5. You can apply emphasis in both, positive and negative prompts.