Mastering realistic photography using Stable Diffusion XL

Introduction

I began using Stable Diffusion in 2022 when AI image tools became very popular. I was excited to create images, but I did not learn the basics of how it works. Instead, I copied prompts and settings from others - a mistake that slowed me down. This article explains the important things I wish I had learned from the start.

Common misconception: Using “Realism” for Realistic images

Many people believe that adding the word “realism” or “photorealistic” to their prompts will produce more realistic-looking images. However, this is actually a significant misunderstanding that can lead to unexpected results.

The terms “realism” and “photorealistic” are primarily associated with art styles rather than actual photography. When you use these words in your prompts, you’re essentially asking the AI to create an artistic interpretation that mimics reality, not to create something that looks like an actual photograph. Avoid these keywords in your prompts if you want to generate realistic photographs.

Positive Prompt

To generate images that look like actual photographs in Stable Diffusion, you should:

• Use photography-specific terminology. Use the keywords like “photo”, “photograph” or “photography”.
• Use keywords related to photography types like “photoshoot”, “editorial”, “stock” or “studio”.
• Add keywords like “studio lighting, “rim lighting”, “short lighting” or “golden hour”. These keywords help create more realistic lighting conditions typical of professional photography.
• Specify camera equipment and film. You can specify camera brands like “Kodak”, “Fujifilm” or “Nikon”. Equipment like “85mm lens” or “f/2.8 aperture”. Film types like “Kodak Portra 160” or “Rollei Infrared 400”.

Negative Prompt

The Negative Prompt tells the AI what it should avoid in the generated image. In case of generating realistic photography, you should specify what you don’t want to see.

• Things that make the image look bad like “low quality”, “low resolution”, “grainy”, “pixelated” or “distorted”.
• Media types like “illustration”, “cartoon”, “painting”, “3D render”, or “artwork”.
• Elements like “text”, “watermark”, “signature”, “error” or “artifact”.

Complete Prompt Examples

Positive:

photo of a young woman with curly brown hair, standing near a window, natural lighting, golden hour, soft shadows, shot with 85mm lens, Fujifilm camera, shallow depth of field, bokeh background, professional photography, high quality

Negative:

illustration, cartoon, painting, 3D render, artwork, low quality, pixelated, grainy, blurry, distorted, text, watermark, signature, unrealistic