Stable Diffusion is a powerful AI system that can generate amazingly realistic images from text prompts. However, writing good prompts that produce quality results takes skill and practice. This article provides prompt examples and tips to help you create realistic images with Stable Diffusion.
What Makes a Realistic Image?
For an AI-generated image to look realistic, it needs to have:
- Sharp focus and detail
- Consistent and natural lighting
- Accurate anatomy and proportions
- Appropriate depth, perspective, and composition
- Photographic qualities like noise, lens effects, etc.
Paying attention to these elements in your prompts will increase realism.
Prompt Structure
Describe the Subject
Start prompts with a detailed description of the subject or main focus of the image. For a person, describe features like age, gender, ethnicity, hair, clothing, etc. For a scene, describe the setting, objects, colors, lighting, etc.
Examples:
- A photo of a 25-year old Caucasian woman with long blonde hair, green eyes, red lipstick, and a blue dress, standing in a forest glen with rays of sunlight streaming through the trees
- A still life photograph of a bowl of lemons and limes with water droplets on a wooden table, soft window lighting, shallow depth of field
Set the Style
Clarify the artistic style. Terms like “photograph”, “portrait”, “cinematic”, etc. trigger the AI to render more realistically.
Examples:
- A professional photograph of a businesswoman in a power suit walking through a modern office lobby. Shallow depth of field, soft office lighting.
- A dramatic side-view portrait of an elderly Native American chief staring into the distance. Beautiful lighting and texture.
Add Photographic Details
Include technical details a photographer would use – lighting, lenses, camera data, post-processing, etc. This greatly improves realism.
Examples:
- A striking close-up portrait of a female model with flawless skin and bright green eyes, shot with an 85mm prime lens at f/2.2 aperture on a full frame camera, post-processed to have a faded film look
- A full body photo of a strong, muscular man on a white background taken with a Canon 5D Mark IV, top down softbox lighting, 85mm lens
Use Negative Prompts
Add “negative prompts” – things you DON’T want – to further refine the image.
Examples:
- A vintage family photo, smiling parents holding a baby, depth of field, 35mm film, vintage border – no modern clothes, no cell phones
- A macro photograph of a red and black poison dart frog in a rainforest, shallow depth of field – not blurry, no other animals
Useful Prompt Keywords
Here are some keywords that generally improve realism in Stable Diffusion prompts:
- 4k, 8k, high resolution, extremely detailed
- depth of field, bokeh, lens flare
- dynamic range, cinematic lighting, rim lighting
- 35mm, 50mm, 85mm lenses
- shallow focus, tack sharp focus
- vibrant colors, rich textures
- Canon, Nikon, Hasselblad cameras
- post-processed, retouched
And useful negative prompt terms:
- low quality, jpeg artifacts, pixelated, blurry
- 3D, CGI, rendered, cartoon, animation
- perfect, idealized
- extra fingers, mutated hands
- textureless, flat, muddy
Prompt Websites
Here are some useful sites for Stable Diffusion prompts:
- Lexica – Searchable gallery of SD images and prompts
- Prompt Hero – Curated prompt gallery
- Promptomania – Prompt generator
Conclusion
With practice, you can learn to write prompts that produce photorealistic images in Stable Diffusion. Pay close attention to details like lighting, depth of field, resolution, and post-processing effects. Refine your prompts with negative prompts. Refer to the examples and keywords in this guide to improve realism. Check the recommended websites for more Stable Diffusion prompts.