Stable Diffusion is a powerful AI system for generating images from text descriptions. With the right prompts, it can create stunning and creative visuals. However, writing good prompts can be challenging for beginners. This article provides a “cheat sheet” of prompt examples, techniques, and formatting tips to help you get the most out of Stable Diffusion.
Prompt Structure
A basic Stable Diffusion prompt has a few key components:
- Description – A detailed description of the desired image content, style, composition etc. Be specific.
- Modifiers – Additional keywords to tweak the output, like “cinematic lighting”, “intricate details” etc.
- Negatives – Words and phrases to exclude from the image.
Here is a prompt template to follow:
[Detailed description of image content and style]
[Modifiers to refine style and quality]
Negative prompt: [List things to exclude]
When starting out, focus more on the description first before adding advanced modifiers and negatives.
Description
The description is the most important part of the prompt. Here are some tips:
- Describe the subject matter, setting, style, composition etc in detail
- Use adjectives and descriptive phrases
- Specify number of subjects and relationship between subjects
- Indicate emotional tone, lighting, atmosphere etc
- Give context and background if needed
Example
A majestic white Bengal tiger resting peacefully under a shady tree in a vibrant green jungle, soft sunlight filtering through the leaves creating a soothing atmosphere, intricate stripes on the fur visible, portrait framing
Modifiers
Modifiers are additional keywords used to refine the output image’s style, quality and details. For example:
intricately detailed, extremely high resolution, photorealistic, National Geographic photography, depth of field
Here are some common categories of modifiers:
Style / Composition – cinematic framing, rule of thirds, aerial shot
Quality / Detail – 8k resolution, ray tracing, Unreal Engine 5 rendering
Lighting / Color – soft lighting, golden hour, high contrast
Medium / Style – impressionist painting, art station concept art
Experiment with modifiers to develop your own style.
Negative Prompt
The negative prompt specifies words and concepts to exclude from the image. For example:
Negative prompt: low quality, bad anatomy, extra limbs, mutation, mutilated, disfigured, poorly drawn hands
Some common negatives:
- Technical faults: low quality, JPEG artifacts, compression errors
- Anatomical issues: malformed limbs, mutated hands
- Unsafe content: nudity, pornography, gore
- Context mismatches: irrelevant objects/settings
Advanced Techniques
Here are some more advanced prompt engineering techniques:
Specify Multiple Samples
Use the :
operator to specify different samples in one prompt.
Example:
A cute baby sea otter floating on its back in the ocean:1.2, under clear blue skies and warm sunlight:1.3, gentle waves lapping around it:1.4
Weight Keywords
Use []
brackets to weight certain keywords and phrases to focus the image on key aspects.
Example:
[Majestic] white Bengal tiger [resting peacefully] under a shady tree in a [vibrant green] jungle
Iterative Prompting
Start with a basic prompt, review the results, and iteratively add details to better match the desired outcome.
Conclusion
With the prompt engineering tips outlined here, you should be able to start crafting prompts that produce stunning images tailored to your creative vision. Remember to start simple, use lots of descriptive details, leverage modifiers for refinement, exclude unwanted elements with negatives, and don’t be afraid to iteratively improve your prompts. The key is practice – have fun with Stable Diffusion!
Useful Resources
- PromptHero – Search engine for Stable Diffusion prompts
- Stable Diffusion Notebook – Jupyter notebook to refine prompts
- PromptMania – Share and discuss prompts