Long read

Prompting Portraits That Hold Up

A portrait is one face carrying the whole image. Pick the frame, light the face, choose one expression, and keep the style from eating the subject.

Free Midjourney Stable Diffusion Chat models 3 min read

One face, one job

Portraits fail when the prompt asks the face to do too much. Two expressions, three style words, and a busy background split the model's attention, and the face is where the seams show. The fix is a budget: one subject clause, one frame word, one light word, one expression, one style. Every worked prompt below fits that budget.

Write the subject as a person with one concrete anchor: an old fisherman with salt-white stubble, a violinist in a velvet jacket. The anchor gives the model something to build the face around. A bare a woman averages every face it knows, and averages look like nobody.

Pick the frame

Frame words set how much of the person is in the picture, and the model reads them before anything else. From tight to wide: extreme close-up, close-up portrait, headshot, bust portrait, half-body portrait, three-quarter portrait, full-body portrait. Lens words do the same job in camera language: 85mm lens flatters faces, 35mm brings in the room, telephoto compresses and blurs the background.

Pick one. Close-up headshot, full body visible is a contradiction, and the model resolves contradictions by ignoring one side. The photography wall holds the full set of framing and lens words.

Light the face

Light does more for a portrait than any style word. The classics each have a name the models know: Rembrandt lighting puts a triangle of light on the shadow cheek. butterfly lighting is the glamour setup, shadow under the nose. rim lighting draws a bright edge and works for silhouettes and drama. window light is the soft documentary default. golden hour warms skin and forgives everything.

Match the light to the mood before you reach for a mood word. A melancholic face under party light argues with itself. The lighting wall sorts the options by what they do.

Expression and body

The emotions wall splits into words the face can show and words the scene can carry. Expression words are literal muscles: furrowed brow, tight-lipped smile, wide-eyed. Mood words color the whole frame: serene, brooding, defiant. Use one of each at most, and let body language words like arms crossed or looking over shoulder replace a second mood word; they read at a distance where facial detail dissolves.

Keep the style in its lane

Style words fight faces. Heavy treatments like cubism or glitch art rebuild the head on their own terms; painterly words like oil painting, gouache, or a film stock like Portra 400 keep the likeness and change the skin. If the person matters, pick from the second group. If the style matters more, let the face go and commit. Half measures produce the uncanny middle.

For a portrait that must stay the same person across many images, that is its own problem with its own tricks; the consistent characters guide covers it.

Worked prompts

window lightclose-up portrait of an old fisherman with salt-white stubble, window light, quiet gaze, Portra 400
rembrandtbust portrait of a violinist in a velvet jacket, Rembrandt lighting, serene, oil painting
rim dramathree-quarter portrait of a boxer in a dark gym, rim lighting, determined, cinematic still
golden hourhalf-body portrait of a potter with clay-dusted hands, golden hour, tight-lipped smile, 85mm lens
storybookheadshot of a young botanist with pressed flowers in her hatband, soft morning light, curious, storybook illustration

When the face breaks

Warped hands, drifting eyes, and melted jewelry are render problems, and they have render fixes: rerun, inpaint, or steer in a chat model with same person, fix the eyes. The fix the picture guide walks the repair loop. In Stable Diffusion, the negative prompt carries the standing face exclusions.

Published here first.