AI models can produce images that look like real photos, but generic prompts often give a glossy, artificial look. The secret is to describe the image the way a photographer would plan the shot.
Say it's a photograph
Start with "A photorealistic photograph of…" or "A candid photo taken on a 35mm film camera of…". This pushes the model toward realism instead of illustration.
Specify camera and lens
- 35mm: natural, documentary feel
- 50mm: close to how our eyes see
- 85mm: flattering portraits with blurred backgrounds
- Macro: extreme close-ups
- Wide-angle 16–24mm: landscapes and interiors
Add aperture for depth of field: f/1.8 for a blurry background, f/8–f/11 for everything sharp.
Describe the light
Light is what makes photos believable. "Soft overcast daylight", "warm golden-hour backlight", "harsh midday sun with strong shadows" or "a single window light from the left" all create very different images.
Ask for imperfections
Real photos aren't perfect. Mention natural skin texture, film grain, slight motion blur, dust, worn materials or a messy background. Avoid words like "flawless" and "perfect", which push toward a plastic look.
Capture a moment
"Candid", "unposed", "mid-laugh", "caught in the moment" make scenes feel alive.
Example
A candid photorealistic photo of a grandfather teaching his granddaughter to ride a bike on a quiet suburban street, late afternoon golden light, shot on a 50mm lens at f/2, natural skin texture, slight motion blur on the wheels, subtle film grain.