I have been experimenting with Stable Diffusion and using it to generate images from text prompts. As an AI enthusiast, I’m fascinated by the creative potential of this new technology. In this article, I’ll share some of my favorite prompt examples and tips for crafting effective prompts.
Crafting Your Prompt
When writing a prompt for Stable Diffusion, there are a few key things to keep in mind:
Specificity
Be as specific as possible in describing the image you want to generate. Details like subject matter, style, composition, color palette etc. will help steer the output.
For example:
A scenic photograph of a winding path through a lush green forest in the Pacific Northwest on a sunny day
Use Adjectives Creatively
Adjectives allow you to tweak attributes like mood, lighting, depth of field and more. Get creative!
A majestic snow leopard with fluffy fur walking proudly on a cliff in the Himalayas
Avoid Ambiguity
Ambiguous prompts increase unpredictability in the AI’s output. Stick to concrete descriptions.
Set the Tone
You can make requests, give suggestions or speak authoritatively. I prefer a more casual, suggestive tone.
For example:
Could you please generate a cute cartoon robot with big eyes, short metal legs and antennae on its head?
Example Prompts
Here are some fun prompt examples I’ve tested with great results:
Stylized Portraits
A stylized oil painting portrait of a girl with blue eyes, dark hair and light skin wearing a red hat
Pop Culture Mashups
Homer Simpson as an astronaut floating through space with the Earth, moon, and stars in the background
Imaginary Creatures
A mystical phoenix with red, orange, and yellow feathers emerging from flames
Landscape Photography
A wide angle photograph of a sunny beach with turquoise water and palm trees bending in ocean breeze
Advanced Prompting Techniques
As you get more advanced, you can start leveraging special techniques:
Seed
The seed allows you to “resurrect” an image you liked and make variations.
For example:
seed: 6378128712, A close up portrait of a cute corgi dog with a red bow tie against a blue background
CFG Scale
This controls how strongly the AI adheres to your prompt vs explores/expands on its own. Lower values stick closer to your prompt.
CFG scale 7: A still life painting by Henri Matisse of a bowl of lemons on a table with a green and white checkered tablecloth background
Sampler
You can indicate a specific algorithm, like DPM Stepping, to use for image generation. This gives you more control.
For example:
Sampler: DPM Stepping, a scenic photograph of Machu Picchu ruins with llamas walking on grass in foreground, by Ryan McGinley
Helpful Resources
To learn more about crafting Stable Diffusion prompts, check out these useful sites: