Writing Prompts for Portraits

A portrait is the least forgiving thing you can generate, because everyone looking at it is an expert on faces. Landscape mistakes read as style. Portrait mistakes read as wrong. That means the levers worth pulling are not the ones people usually reach for — the mood words matter far less than where the light is coming from and how far away the camera is standing.

Distance Decides the Face

Focal length changes the shape of a head more than any adjective will. A wide lens close to a subject pushes the nose forward and stretches the sides of the face; that is why phone selfies look the way they do. Somewhere around 85mm to 135mm the proportions settle into what we read as flattering, which is why portrait photographers live there. Say it plainly — “85mm portrait lens, waist-up” — and you have fixed both the geometry of the face and roughly how much of the body is in frame.

Crop is a separate decision and worth making consciously. A tight head-and-shoulders crop is about the person; a three-quarter or full-length frame is about the person in a place, and the moment you go that wide the clothes, the posture and the room all become part of the subject and need describing. Half of the bad portraits I see are wide frames written as if they were close-ups, so the model invents the trousers and the floor.

Light Direction Is the Whole Mood

Naming the light source and its direction does more than any lighting adjective. Soft window light from the left gives you a gentle falloff across the cheek and a visible catchlight. Hard direct sun overhead gives you sunken eye sockets. A single lamp low and to one side gives you drama and a hard shadow line down the nose. Backlight at the end of the day gives you a rim on the hair and a face that needs fill from somewhere or it goes dark.

The useful habit is to write light as a position, not a feeling. “Warm, moody lighting” gets interpreted anywhere; “low golden sun from behind and slightly right, face lit by a bounce” gets built. Pair that with the surface — overcast sky, softbox, bare bulb, candle — and the quality of the shadow edge follows automatically.

Give the Face Something to Do

Expression collapses to a neutral half-smile unless you write otherwise, and a neutral half-smile is the visual signature of a generated portrait. Describe an action instead of an emotion: looking off past the camera, mid-laugh with the eyes closed, chin down and eyes up, glancing back over the shoulder. Where the eyes go changes the entire read of the picture, and eye direction is one of the instructions models follow reliably.

Age, build and clothing deserve one specific phrase each rather than a pile of adjectives. “A woman in her sixties, short grey hair, in a worn denim jacket” is a person. “Beautiful elegant woman” is a stock photo of nobody.

A quick test before you generate: cover the style words in your prompt and see if what remains would let a photographer set up the shot. If they would still have to ask where the light is and how close to stand, those are the two clauses to add next.