Stable Diffusion is a powerful AI image generation model that can create highly realistic and detailed images from text prompts. However, writing good prompts that produce photorealistic results requires some skill and practice.
In this article, I will provide useful prompt engineering tips and examples for generating photorealistic images with Stable Diffusion. Whether you’re a beginner just starting out or looking to improve your prompt writing, this guide has prompt ideas to spark your creativity.
Specify Camera and Lens Details
One of the most important things you can add to a Stable Diffusion prompt for realism is specific camera and lens information. This gives the AI clearer instructions on the type of image you want.
Here’s an example prompt structure:
“A beautiful portrait of a woman, photographed with a Canon 5D Mark IV camera using an 85mm f/1.2 lens, shallow depth of field, intricate lighting”
Specifying details like the camera model, lens details, lighting, and depth of field results in more realistic and visually pleasing portraits.
You can use this for any type of scene – just replace the subject and add relevant camera specs.
Use Photographic Style Keywords
Along with camera details, using specific photographic style keywords also improves realism.
Some examples:
- Cinematic
- Dramatic
- Film noir
- High key
- Low key
- Soft focus
- Shallow depth of field
- Bokeh
For example:
“A romantic couple walking in the rain, cinematic style, shallow depth of field, soft bokeh lights”
These terms nudge the AI to apply those stylistic rendering techniques for enhanced realism.
Include Studio Lighting Setups
You can suggest specific studio lighting arrangements like:
- Rembrandt
- Loop
- Split
- Rim
Using these along with modifiers such as softbox, umbrella, and snoot will result in more accurate studio portrait lighting.
For example:
“A female model posing in the studio, dramatic Rembrandt lighting, softbox placed above and left”
Use Negative Prompts
Negative prompts are terms and concepts you want the AI to avoid. Using these eliminates unwanted elements that reduce realism.
Some examples:
- cartoon
- CGI
- fake
- perfect
- extra fingers
- mutated hands
- poorly drawn
For example:
“A couple standing below a oak tree on a fall day, photorealistic, high resolution, -cartoon, -CGI”
The negative prompts instruct Stable Diffusion to avoid generating cartoon or CGI elements.
Mention Photo Manipulation Apps
You can also name specific photo manipulation apps to suggest the type of processing you want:
- Photoshopped
- Lightroomed
- Filmic pro
- Retouched
- Color graded
For example:
“A macro photography of a red rose, Photoshopped, 8K resolution”
This prompts Stable Diffusion to apply realistic photo retouching and post-processing effects.
Use Landscape Photography Keywords
For outdoor and landscape images, using specialized landscape photography terms goes a long way in improving realism.
Some examples:
- Golden hour
- Blue hour
- Sunrise
- Sunset
- Hazy
- Silhouette
- Long exposure
- Tilt shift
Here’s a sample prompt with some of these terms:
“Majestic mountain landscape during golden hour, long exposure of river, tilt-shift effect”
Including niche landscape concepts like these prompts Stable Diffusion to render lighting and camera effects that match real landscape photos.
Prompt Formatting Tips
Here are some key prompt formatting best practices for better photorealism:
Keep Prompts Concise
Don’t overload your prompts with too many details and modifiers. Concise prompts around 30-50 words tend to produce better images.
Order Details from Broad to Specific
Structure your prompts to first describe the broad scene, subject matter, and style. Then go into more nuanced lighting, camera, and post-processing details.
Separate Concept Sets with Commas
Use commas to separate logical “concept sets” related to scene, subject, lighting, camera, etc. This improves clarity.
Capitalize Photo Terms
Consider capitalizing important photo-related terms like lighting arrangements and camera brands. This emphasizes them.
Use Present Tense
Describe the scene and subject matter primarily in the present tense for clarity. For example, “A woman stands in a forest…”
Helpful Prompt Engineering Websites
Here are some useful websites for crafting better Stable Diffusion prompts:
- PromptHero – Browse and search a huge database of prompts for inspiration
- Nightcafe Creator – Get AI-generated recommendations to improve your prompts
- Automatic1111 WebUI – Advanced features like image embeddings and prompt weighting
These tools make it easier to fine-tune prompts for improved photorealism.
Useful Prompt Engineering Techniques
When crafting prompts, also experiment with techniques like:
- Prompt weighting – Increase influence of key terms
- Image embeddings – Base prompts on sample images
- Prompt sequences – Chain prompts on new concepts
Adjusting these advanced parameters can noticeably enhance image quality.
Example Photorealistic Prompts
Here are a few full example prompts for generating photorealistic images with Stable Diffusion:
- “A golden retriever puppy sitting in grass, photographed with a Sony A7R IV camera using 100mm macro lens, shallow depth of field, soft natural lighting, high resolution and detail, commercial product shot”
- “Majestic snowy mountain sunset landscape, photographed with a Nikon D850 using 16-35mm wide angle lens, cinematic style, long exposure silhouetted trees, vivid colors, dynamic range”
- “Two young women laughing together in a trendy cafe, photojournalistic style, Canon 5D Mark III camera, 50mm lens, soft bokeh background, shaded window light, color graded in Lightroom”
As you can see, using the right prompt structure and photographic keywords results in much more realistic outputs from Stable Diffusion.
Follow these prompt engineering best practices, and experiment with the parameters. With some practice, you’ll be able to create photorealistic images ready for print, websites, and commercial use.
Conclusion
- Use specific camera and lens details
- Include photographic style keywords
- Suggest studio lighting arrangements
- Leverage negative prompts
- Mention photo editing apps
- Use niche landscape terms
- Format prompts concisely and logically
- Capitalize photo concepts
- Browse prompt sites for inspiration
- Adjust advanced parameters like weighting
Carefully engineered prompts that describe a realistic photographic scene allow Stable Diffusion to render images that resemble real-world photographs. With the prompt engineering tips outlined in this article, you can guide the AI to produce professional, print-worthy images.
References
Automatic1111 Stable Diffusion Web UI
Nightcafe Creator
PromptHero