AI prompts for portraits and headshots that do not look AI made
Lighting, lens, pose, and skin are where the AI face gives itself away. The realism levers for portraits, how to avoid the plastic look, and prompt structures for clean headshots.
The core idea in one paragraph
Portraits are where AI image generation gives itself away most obviously, because the human brain is exquisitely tuned to faces and catches even small inconsistencies instantly. The plastic, too perfect look that everyone associates with AI images is overwhelmingly a portrait problem, and it comes from leaving the skin, the light, and the lens to the model's defaults. The model's default face is idealized, symmetrical, and evenly lit, which is exactly the combination the brain reads as fake. The fix is to understand which levers push a portrait toward real: flattering but specific lighting, a lens that reproduces natural perspective, skin that shows texture and imperfection, and a subject with character rather than idealized perfection. Pull those levers deliberately and your portraits stop looking generated.
What this article covers
The lighting setups that flatter faces, the lenses that render them naturally, the skin and imperfection cues that defeat the plastic look, and prompt structures for both creative portraits and clean professional headshots.
Why the face is the hardest test
Before the techniques, appreciate why portraits are so unforgiving, because the reason points directly at the solution. The human visual system devotes a disproportionate amount of its processing to faces. We are built to read expression, identity, and intent from faces instantly and subconsciously, which means we detect even tiny inconsistencies in a rendered face that we would completely miss in a landscape or a still life. A tree with one too many branches does not bother us. A face with subtly wrong skin texture or asymmetrical lighting bothers us immediately, even if we cannot articulate why.
This extreme sensitivity is why the model's defaults fail so visibly on faces. The default portrait the model produces is a statistical average of all the portraits in its training data, which trends toward idealized, symmetrical, evenly lit, and smooth skinned. That average is real in the sense that many training images share those traits, but it is fake in the sense that almost no real face is all of those things at once. Real faces have asymmetry, texture, imperfection, and they are lit by real light with direction and shadow. The gap between the model's average and a real face is exactly what the brain flags as wrong.
The lighting that flatters and reads as real
Light is the most powerful lever in portrait work, and it is where the model's default fails most. The default portrait light is even, frontal, and shadowless, which flattens the face and removes the dimensionality that real portrait photography depends on. Real portrait light comes from a specific direction, creates gentle shadows that describe the form, and reveals the three dimensional structure of the face.
The classical portrait lighting setups, borrowed from centuries of painting and photography, are the most reliable cues because the model has extensive, consistent training data for each. Loop lighting, with a soft side light at a gentle angle casting a small nose shadow toward the cheek, is the most universally flattering and the safest default for almost any face. Rembrandt lighting, with a stronger side light creating a triangle of light on the shadowed cheek, is more dramatic and sculptural. Butterfly lighting, with light from above and slightly in front casting a small shadow under the nose, is glamorous and classic. Each of these adds dimension that the flat default lacks.
Beyond the setup, describe the light's quality and direction explicitly. Soft window light from the left is real and flattering because it names a real source at a real angle. Golden hour side light adds warmth that reads as health. Avoid hard overhead light, which carves shadows under the eyes and ages the subject, and avoid flat frontal light, which erases the form. The right light is the single biggest difference between a portrait that reads as real and one that reads as generated.
The lens that renders a face naturally
Lens choice is the second lever, and it matters more in portraits than in almost any other subject, because the wrong lens distorts the face in ways the brain detects instantly. The model's default lens tends toward a slightly wide perspective, which exaggerates features close to the camera and produces the subtly wrong proportions that read as off.
The classical portrait lens is a short telephoto, in the 85mm to 105mm range, which compresses the face gently, reduces distortion, and reproduces the proportions the brain expects from a real photograph. This is why portrait photographers default to 85mm and 105mm lenses, and it is why naming those focal lengths in a prompt produces more natural faces. Pair the lens with shallow depth of field, so the eyes are sharp and the ears soften, which is the signature look of real portrait photography and a strong authenticity cue.
For environmental portraits that include more context, a 50mm normal lens gives a slightly wider view with still natural perspective. Avoid wide angle lenses for close portraits, because the distortion they introduce on faces reads as a mistake unless you are intentionally going for an unusual effect. The lens is not a minor detail in portrait work. It is a primary determinant of whether the face looks right.
Skin and the defeat of the plastic look
This is the lever that does the most work to defeat the AI face, and it is the one people resist most, because it feels counterintuitive. The plastic look comes from skin that is too smooth, too even, and too perfect, which is the model's default because its training data includes heavily retouched photography. Real skin has texture: pores, fine lines, slight color variation, imperfections. Reproducing that texture is what makes a face read as captured rather than rendered.
Add skin texture and imperfection deliberately. Visible skin texture, fine pores, slight asymmetry, flyaway hair, natural skin tone variation are all cues that defeat the smooth default. A real face also has character: age, expression lines, the marks of a life lived. Specifying age and letting the face show it produces far more convincing portraits than asking for youth and beauty, because idealized youth is exactly where the model's average lives and where the fake detector fires hardest.
The character principle
The most convincing AI portraits are of faces with character, not faces with perfection. An older face with lines and texture reads as real almost immediately. A young idealized face reads as fake almost immediately. If you want realism, lean into specificity, age, and imperfection, and away from the model's default of beauty.
Expression, pose, and the eyes
Beyond light, lens, and skin, the expression and pose carry enormous weight in how a portrait reads, and they are often left vague. A default expression from the model tends toward a blank, slightly unfocused stare that is one of the most reliable tells of a generated face. Real portraits have intention in the expression and the pose, and specifying that intention is a powerful lever.
Describe the expression precisely. A natural relaxed almost smile beats smiling, because the model's default smile is too wide and too symmetrical, while an almost smile reads as genuine. A thoughtful, faraway gaze beats looking at the camera, because a specific gaze has intention where a default gaze does not. The eyes matter especially, because the viewer looks at the eyes first and longest, and eyes with a clear point of focus and a hint of expression carry the portrait.
Pose and body language add the rest. The angle of the head, the set of the shoulders, the position of the hands if visible, all of these communicate character and mood. A three quarter head turn with relaxed shoulders reads as approachable. A direct front facing pose with squared shoulders reads as confident. A slight tilt and a hand near the face reads as intimate or thoughtful. Specify the pose, because the model's default pose is stiff and symmetrical, and specific posing is where a portrait gains life.
Clean headshots for professional use
Creative portraits and professional headshots share the lighting and lens logic, but headshots have a different goal: a clean, trustworthy, neutral image suitable for business contexts. The headshot is the corporate and personal branding staple, and AI generation is excellent for it once you understand the conventions.
A professional headshot uses simple, flattering light, a neutral background, and a natural expression. The light is usually soft loop or butterfly lighting from slightly above, which flatters without drama. The background is a solid neutral color, often grey, blue, or white, with no distraction. The expression is a natural, relaxed almost smile, not a grin and not a blank stare. The lens is the standard 85mm to 105mm portrait telephoto, with shallow depth of field that keeps the face sharp and the background soft.
Here is a prompt structure for a clean professional headshot:
[PERSON DESCRIPTION with age and defining features], wearing [professional attire], soft loop lighting from above left, neutral [grey] seamless background, shot on an 85mm lens at f2.8, shallow depth of field, natural relaxed expression, visible skin texture, professional corporate headshot photography, color accurate
For consistent headshots across a team, lock the lighting, background, lens, and style description, and vary only the person description, exactly as in the batch product system. A set of team headshots generated within a locked system looks like one photographer shot them all, which is exactly what a professional team page needs.
Creative portraits with mood and story
Creative portraits have more freedom, and they are where you can use the lighting and mood vocabulary to full effect. The goal shifts from neutral trust to emotional impact, and the choices become more expressive. The lighting setups become more dramatic: low key for mood and mystery, Rembrandt for sculptural drama, mixed practical light for realism and story. The lens may stay telephoto for flattering compression, or move wider for environmental portraits that place the subject in a meaningful context.
For a moody creative portrait, a structure like this works:
[PERSON DESCRIPTION], in [a dimly lit workshop at night], low key lighting with a single warm practical light from the right, deep shadows on the left side of the face, shot on an 85mm lens, shallow depth of field, visible skin texture and character, documentary portrait photography, intimate and quiet mood
The creative portrait benefits from the full mood engineering approach covered in the light and mood article, where direction, hardness, and color are chosen to produce a specific feeling. Portraits are especially responsive to mood engineering, because the face carries emotion so strongly that the right light on the right expression produces immediate emotional impact.
Avoiding the common portrait failures
A few habits consistently produce fake looking portraits, and avoiding them does most of the work. The first is leaving the light to the default, which produces flat, shadowless faces that read as plastic. Always specify a directional portrait light. The second is using a wide or default lens, which distorts proportions. Always name a portrait telephoto focal length. The third is idealizing the skin and features, which lands squarely in the model's average where the fake detector is most sensitive. Add texture, age, and character deliberately. The fourth is excessive smoothing in any post processing, which reverses all the texture work you did in the prompt. Preserve the imperfection through the whole pipeline.
The fifth and most important is treating the face as a generic subject rather than a specific person. Generic faces are exactly what the model produces by default, and they are exactly what reads as fake. The most powerful realism lever in portrait work is specificity: a particular age, a particular feature, a particular expression, a particular life. The more specific the person, the more the model reaches past its average into the territory where real faces live, and the more convincing the portrait becomes. Describe a person, not a face, and the result follows.
