Prompting for Portraits: People, Characters, and Faces

Faces are the hardest thing a diffusion model can render and the most rewarding to get right. A well-written portrait prompt produces a believable, characterful person you can almost imagine meeting. A vague one produces the dreaded "AI face" — symmetrical, plasticky, indistinct. This tutorial gives you the deliberate framework for portraits that look like real people.

1. Introduction

Portraits demand more from a prompt than almost any other genre because viewers know exactly what a human face is supposed to look like. The slightest off detail — uncanny eyes, melted ears, glossy plastic skin — pulls the viewer out instantly. The fix is not "more detail keywords". It is a deliberate description of who this person is, what they look like, how they feel, and how they are framed.

This tutorial covers prompts for three portrait types — real-feeling people, stylised characters, and headshots — and shows you the vocabulary that works for each.

2. The Concept Explained

Every good portrait prompt answers five questions: who is this person, how do they look (physical), what are they wearing, how do they feel (expression and posture), and how are they framed (camera and lighting). Skip any of these and the model invents — and its inventions tend to converge on a bland, symmetric, slightly glossy "AI default" face.

Think of it like commissioning a painter. You would not say "paint my friend". You would describe her: late forties, freckles, slight gap in her front teeth, wearing the green wool coat she always wears, laughing mid-sentence the way she does, sitting on the kitchen step in afternoon light. That level of specificity is what models also need.

Identity and physical description

Useful elements: age range (early thirties, mid-fifties), ethnicity and heritage, body type, hair colour and style, eye colour, skin texture (freckles, lines, sun-weathered), distinctive features (gap in teeth, scar above eyebrow, asymmetric smile, dimples). Asymmetry is your friend — symmetric prompts produce that uncanny "AI face".

Wardrobe and styling

Specify the era, fabric, and condition. "Worn green wool overcoat with a frayed collar" is far more evocative than "a green coat". Wardrobe carries an enormous amount of character — a person's clothes communicate class, era, profession, and mood in a single phrase.

Expression and posture

Faces feel alive only when the expression is named. Useful: warm half-smile, gentle laugh, contemplative gaze off-camera, serious focused expression, tired but kind, mid-conversation, looking directly at the camera with quiet confidence. Add posture: shoulders relaxed, leaning slightly forward, hands clasped in lap.

Camera framing

For portraits, the most common useful options are: extreme close-up (face fills frame), tight portrait crop (head and shoulders), waist-up environmental portrait, full-body portrait. Lens choice matters here: an 85mm portrait lens compresses pleasantly and is the default for flattering headshots; a 35mm lens feels more documentary and intimate.

3. The Problem Without This Structure

portrait of a beautiful woman

4. The Solution

Environmental portrait of a 62-year-old Portuguese
fisherwoman, weathered sun-tanned skin with deep
laughter lines around her eyes, short steel-grey hair
tucked under a faded blue cotton headscarf, one
asymmetric gold tooth visible when she half-smiles.

Wearing a worn navy fisherman's sweater patched at
the elbow, a small silver crucifix at her neck.

Standing on a stone harbour wall at dawn, arms folded,
looking directly at the camera with quiet pride.

Lighting: soft blue-hour ambient light from the sea
behind her, a warm rim from the harbour lamp on her
left cheek. 35mm documentary lens, slight film grain.

Style: photorealistic, shot on Leica Q3, in the spirit
of Steve McCurry documentary portraiture.

--ar 4:5 --v 6 --style raw

5. Step-by-Step Breakdown

1. Decide who they are first. Profession, age, heritage. Identity gives the model a vector to start from.
2. Add one or two asymmetric features. A small scar, a gap in teeth, sun damage on one side of the face, a tooth that catches the light. Asymmetry is the antidote to "AI face".
3. Dress them specifically. Material, era, condition. "Frayed collar", "patched elbow", "linen apron faded with use".
4. Direct the expression. Tell the model what they feel and where they are looking. "Quiet pride", "mid-laugh", "looking off-camera, lost in thought".
5. Frame the shot. Choose a portrait crop, name a portrait lens (50mm or 85mm), and place the light. Soft side-light or a warm rim is almost always flattering.

Tip: Avoid the word "beautiful". It pulls models towards their plastic, symmetric ideal. Replace it with the specific feature you actually find beautiful — "kind eyes with deep laughter lines", "soft natural smile".

6. Practice Exercises

7. Key Takeaways