Stable Diffusion Prompting Guide

Understand that Stable Diffusion is a deep learning model designed to generate images from text descriptions, capable of inpainting, outpainting, and image-to-image translations.

Create detailed prompts by incorporating keyword categories such as subject, medium, style, artist, website, resolution, color, and lighting to enhance image quality and specificity.

Include popular keywords like "digital painting," "portrait," "concept art," "hyperrealistic," and "pop-art," and reference specific artists or websites (e.g., Artstation, DeviantArt) to influence style.

Use syntax to adjust keyword importance (weight) in tools like the AUTOMATIC1111 GUI, such as (keyword: factor), parentheses (), or double curly braces {{}} for emphasis adjustments.

Apply keyword blending or prompt scheduling using syntax like keyword1: keyword2: factor to create smooth style transitions during image generation.

Be mindful of token limits (about 75 tokens per prompt in basic models) and consider association effects where attributes may influence other image elements like ethnicity or pose.

Iteratively refine prompts by starting with simple subject, medium, and style, then gradually add keywords to improve output quality.

Use universal negative prompts to exclude unwanted elements from generated images.

Recognize that early diffusion steps determine overall image composition, while keyword adjustments later influence finer details.