Skip to main content
All guides
Portrait photographyBy Prompt TemplateVerified on GPT Image 2 / Nano Banana

Write f/1.4, not “shallow depth of field”

Portraits look like CGI mannequins until you use camera language and imperfect skin. 50mm and f/1.4 beat “half-body, cinematic,” and pores beat “beautiful.”

2026/09/01

Image models respond to camera notes more than adjectives. “Cinematic half-body, shallow depth of field, beautiful woman” is how you get a wax figure in a smeared bokeh. Sony A7R V, 50mm, f/1.4, Kodak Portra 400, visible pores is how you get a photograph.

What actually breaks

We generated “portrait of a barista for a Prompt Template case study, realistic.” Both models delivered poreless skin, jewelry-ad lighting, and a 35mm-ish wide face. Adding “more realistic” added extra eyelashes. The shot only landed when we replaced the adjectives with gear, aperture, film, and defects.

Street “candid” shots fail when you say “real” but describe a studio. Write the mess: wet asphalt, a crooked horizon, a hand-held roll.

For try-on or beauty matrices, skip this guide’s camera slot at your peril — but the fatal miss is still a new face per tile. That lives in the face slot: one identity, only the variable changes.

The rule we use

  1. Lens instead of crop. 50mm ≈ half-body; 85mm ≈ tight portrait; 35mm ≈ environment. Do not write “half-body photo” and hope.
  2. Aperture instead of “blur.” f/1.4 or f/2. “Shallow DOF” is a wish.
  3. Put flaws in the spec. Pores, peach fuzz, a sleep crease, grain. Perfect skin reads as 3D.
  4. Light has a direction. Window left, neon behind, overhead fluorescent — pick one.

Example we actually ran

Weak

Realistic cinematic portrait of a tired barista, shallow depth of field, beautiful lighting.

Locked

Shot on Sony A7R V, 50mm lens, f/1.4, eye-level, handheld.
Subject: 32-year-old barista, denim apron, wiping a ceramic "Prompt Template" mug.
Skin: visible pores, faint under-eye, no beauty-filter smoothing.
Setting: dim cafe, rain on the window behind, tungsten interior vs cool street light.
Film: Kodak Portra 400 emulation, mild grain. Not CGI, not fashion retouch.

GPT Image 2 kept the 50mm compression when 50mm sat in the camera line, not buried after the wardrobe. Nano Banana still over-smoothed until no beauty filter was in the last sentence. Remove the mug brand if you only need the lighting lesson — it is there so we can tell our take from a generic cafe stock prompt.

When to open this guide from a control

camera and aperture are not flavor. They are the crop and the blur. If you change camera from 50mm to 35mm, expect more environment and a rounder face — that is the point of the slot. Keep face locked when the same person has to survive a lighting or wardrobe swap.

Sources

Rewritten from public pitfall notes; the barista + mug portrait is a Prompt Template example.

Controls this guide covers

  • camera

    Use lens language (`50mm`, `f/1.4`) instead of “half-body, shallow DOF”.

  • aperture

    `f/1.4` beats “blur the background”. Models follow camera notes more than adjectives.

  • face

    Lock one face across a grid. Unlocked palettes mint a new person per swatch.

Related templates

Browse Portrait & Photography in the gallery