Getting Realistic Skin in AI Images

TL;DR: Realistic skin in AI images comes from four things working together: visible texture (pores, fine hair, subsurface glow), a specular highlight instead of matte flatness, color variation across the face, and lighting that reveals rather than erases all of it. Below: why the default output skips every one of these, plus 25+ copy-paste prompts.

One thing before any of this: Prompt Architects writes the prompt, not the image. Nothing on this page renders a picture; these are words to paste into whichever model does.

Two Rules That Come Before Any Skin Prompt

Getting either wrong isn't a craft mistake; it's a different kind of problem.

Never generate a real, identifiable person. Not a named celebrity, a specific coworker, or a real historical figure described precisely enough to identify. That isn't a photograph of someone; it's a fabricated likeness, and presenting it as real is exactly what the platforms rule out. Midjourney's Community Guidelines: "Do not intentionally mislead recipients of generated images about their nature or source" (docs.midjourney.com, updated 17 August 2026). The fix costs nothing: describe the person you need instead of naming one. Every prompt below does that.

Skin tone is not a difficulty setting. It isn't a "hard case" to hedge around while treating one tone as the unstated default. Lighting a range of skin tones well is established photographic craft: meter for the person actually in front of the camera. Full section below, with real vocabulary, not euphemism.

Why Does AI Skin Still Look Like Plastic?

Because the default output is the statistical average of the training data, and the average face skews smooth, evenly lit, and retouched. Unless you name texture, you get the average: no pores, no fine hair, no depth under the surface.

Three things do the work when skin reads as real instead of rendered. Pores: finest on the nose and forehead, coarser near the jaw. Fine hair, the nearly invisible down called vellus hair, catches light along the jaw, upper lip, and temples, and is one of the fastest tells a viewer's eye registers without noticing why. Subsurface variation is light entering translucent skin, scattering under the surface, and exiting elsewhere, giving thin areas like ears, nostrils, and eyelids a soft glow instead of a flat, opaque surface.

OpenAI's own cookbook states this as a rule, not just an example: "Use photography language (lens, lighting, framing) and explicitly ask for real texture (pores, wrinkles, fabric wear, imperfections). Avoid words that imply studio polish or staging" (developers.openai.com, GPT Image prompting guide, section 4.3, accessed September 2026). Its worked example reads: "He has weathered skin with visible wrinkles, pores, and sun texture... The image should feel honest and unposed, with real skin texture, worn materials, and everyday detail. No glamorization, no heavy retouching." That last line is the fix in one sentence: ask the model not to smooth it.

Why Does a Matte Face Look Dead?

Because real skin is never fully matte. It has an oil layer and a moisture layer, both of which catch light and throw back a small, soft specular highlight, usually strongest on the forehead, nose bridge, cheekbones, and chin, the parts photographers call the T-zone. A face with zero specular response reads as a painted surface or a mask, not skin, because your eye expects that tiny reflected point and its absence registers as wrong even when nothing else looks off.

The failure runs both directions. Overdo the highlight and skin looks oily or sweaty rather than alive. The target is a small, soft, low-contrast highlight, not a hard glossy one: describe it as "subtle natural sheen across the forehead and nose bridge" rather than "shiny skin," which tends to overshoot into wet-look territory.

Why Isn't Skin One Color Across a Face?

Because it isn't, on a real face, ever. Blood vessels sit closer to the surface at the nose, cheeks, and ears, so those areas read warmer and redder. The jaw and forehead, with thicker skin and less surface vasculature, read cooler and more neutral by comparison. A face rendered as one uniform tone, no matter how accurate that single tone is, looks airbrushed because real skin doesn't hold still like that.

Naming the variation directly works better than naming a single color: "warmer, slightly flushed tone at the nose and cheeks, cooler and more neutral along the jaw" gives the model a real photographic pattern to reproduce, instead of one flat swatch to paint with.

How Does Lighting Decide Whether Skin Texture Shows?

Because texture is a lighting effect before it's anything else. Every pore, every fine hair, every subtle bump is a tiny piece of surface geometry, and geometry only becomes visible when light rakes across it at an angle and casts a shadow. Flat, frontal, diffused light hides almost all of that; light with direction and hardness reveals it.

Ask for harder, more directional light when the prompt needs visible texture; ask for soft, wrapping light for a smoother beauty look, and accept that some texture goes with it. For the full vocabulary, see lighting terms for AI image prompts; this post only covers what lighting does specifically to skin.

Which Imperfections Make a Face Look Real?

The ones a retoucher would normally remove. Freckles, faint acne scarring, a slightly asymmetric brow, uneven tone across the cheeks, fine lines at the eyes: these are exactly what an "enhance" or "beautify" pass strips out, which means they're also exactly what's missing from a model's default smoothed output.

The craft is restraint, not maximum imperfection. One or two specific, located details ("a scatter of light freckles across the nose and upper cheeks" or "a faint asymmetry to the smile") read as a real, specific person. A dozen stacked flaws reads as a deliberate "ugly" filter working just as hard, in the opposite direction, as the airbrushed default it's reacting against.

Why Does Skin Fall Apart in Shadow?

Because shadow strips away the two things skin needs to read as skin: the specular highlight and the subsurface glow both depend on light actually reaching the surface, and a shadow, by definition, is where it doesn't. Flatten the fill light in your prompt and shadowed skin turns matte, waxy, and uniformly dark, losing texture and color variation at once, which is why faces in dramatic AI-generated shadow so often look like a mask rather than a person standing in low light.

The fix is a soft fill, not more key light. A dim bounce, a hint of ambient light, even reflected light off a nearby surface, is usually enough to keep a faint highlight and a trace of subsurface warmth alive in the shadowed side of a face without flattening the shadow itself. "Soft fill light preserving subtle skin detail in the shadowed half of the face" does more useful work in a prompt than "well-lit," which tends to just wash the shadow out entirely.

Makeup vs. Bare Skin: Are You Prompting for the Same Target?

No, and mixing the two vocabularies is a common way prompts fight themselves. Bare, unretouched skin is the texture-forward target this whole post has been describing: visible pores, fine hair, natural color variation, a soft rather than glassy highlight. Makeup is a deliberate, controlled modification of exactly those things: foundation evens tone and can mute pore visibility, powder cuts specular shine down to almost nothing, and contouring adds artificial shadow and highlight in places real anatomy wouldn't put them on its own.

Neither is more "realistic" than the other; they're different targets. A prompt for "bare skin, visible pores and natural texture, minimal makeup" and a prompt for "full glam makeup, matte foundation, contoured cheekbones, soft-focus finish" are both asking for a real photographic look, just two different real looks. Trouble starts when a prompt asks for both at once: "flawless poreless skin" alongside "visible skin texture" gives the model contradictory instructions, and it will pick one, usually the smoother one, since that's still the deeper default.

Why Does Near-Perfect Skin Look Worse Than Stylized?

Because near-perfect is where the uncanny valley actually lives. A clearly stylized illustration doesn't trigger discomfort, because nobody expects it to be a photograph. A rendering that gets 95% of real skin right, wrong in one small, hard-to-name way (a highlight sitting slightly off, a texture a fraction too uniform, proportions technically flawless in a way real faces never are) unsettles more than a bad render or an honest illustration, precisely because the viewer's brain flags the mismatch without being able to explain it. The closer the attempt, the more that one remaining error stands out.

Which is also why a generated face is not a photograph of anyone, and must never be presented as one. OpenAI's own documentation concedes the model can't fully guarantee its side of this: under GPT Image's "Limitations" heading, it states that "the model may occasionally struggle to maintain visual consistency for recurring characters or brand elements across multiple generations" (developers.openai.com/api/docs/guides/image-generation, accessed September 2026; worth re-checking before you quote it, since limitations pages get revised). If a model can't reliably reproduce one invented character twice, it has no business being asked to reproduce a real one once.

How Do You Prompt a Specific Skin Tone Respectfully?

By describing it in the same ordinary photographic vocabulary you'd use for hair color, eye color, or wardrobe, and lighting it the way a photographer would meter for the person actually standing in front of the camera. That's the whole rule. Skin tone is a physical attribute in a portrait prompt, not a topic to talk around.

Concretely: name the tone and its undertone together. Undertone (warm, cool, neutral, and the specific cast: gold, pink, olive, red, blue) changes how light behaves on skin more than the base depth of the tone does. "Deep brown skin with warm, golden undertones" and "fair skin with cool, pink undertones" are both precise, both respectful, and both give the model something concrete to render, which vague language never does. Avoid euphemistic color words as stand-ins for skin: calling a tone "flesh-colored" or "nude" quietly assumes everyone's skin is that one shade.

Lighting matters as much as the color word. A setup built around one tone and applied unthinkingly to another produces two visible failures: deeper tones losing subsurface glow and going flat and dark, or lighter tones blowing out to a washed, textureless white. Both are lighting failures, not skin failures. Meter and describe the light for the actual tone in the shot: "soft key light exposed for deep brown skin, holding detail in the shadow side" reads very differently to a model than "well-lit portrait" applied blind to the same face.

Last: this post makes no claim about whether any specific model renders one skin tone more accurately than another. That number circulates, unsourced, in forum threads and roundups, and without a vendor's own benchmark behind it, repeating it here would be exactly the unverified claim this post argues against making. A real difference you find in your own testing is worth writing down; it isn't one this post will assert for you.

Which Prompts Actually Produce Realistic Skin?

33 copy-paste prompts below, grouped by what they're testing. Every bracket is yours to fill in; everything outside brackets is the vocabulary doing the actual work. These are plain descriptive prompts, meant to work across GPT Image, Midjourney, FLUX.2, Nano Banana Pro, Ideogram, and similar tools; where a parameter is tool-specific, it's noted as an optional add-on rather than baked into the prompt itself.

Texture: pores, fine hair, subsurface glow (4)

Photorealistic portrait of [a specific person: age range, build, hair, expression], visible skin pores and natural fine facial hair catching the light along the jaw and upper lip, subtle subsurface glow at the ears and nose tip, unretouched, honest skin texture, no smoothing or beautification

Close-up portrait, [SETTING], skin showing natural pore texture across the nose and forehead, faint vellus hair visible at the temples in raking light, soft warmth under the skin at the ear rims, shot like a real photograph, not a painting or 3D render

[SUBJECT], candid unposed portrait, real skin texture including pores, fine lines, and natural oil sheen at the T-zone, subsurface scattering visible where light passes through the thin skin of the ears and nostrils, natural color balance, no glamorization, no heavy retouching

Macro-adjacent portrait crop on the cheek and eye area of [SUBJECT], visible skin texture: pore detail, one or two fine lines, subtle color variation, soft natural sheen, avoiding any airbrushed or plastic smoothness

Specular highlights: a face that isn't matte (3)

Portrait of [SUBJECT] under [LIGHT SOURCE], subtle natural sheen across the forehead, nose bridge, and chin, soft small specular highlights rather than a glossy or oily look, skin otherwise matte with visible texture

[SUBJECT] lit by a single soft key light, gentle catchlight in the eyes and a faint natural highlight along the cheekbones and nose, skin reading dimensional rather than flat or painted

Editorial portrait, [SUBJECT], controlled specular highlight at the T-zone only, rest of the face matte with visible pore texture, avoiding an all-over shiny or waxy finish

Color variation across the face (3)

Portrait of [SUBJECT], warmer and slightly flushed skin tone at the nose, cheeks, and ears, cooler and more neutral tone along the jaw and forehead, natural, unretouched color variation, no single flat skin tone

[SUBJECT] in natural daylight, subtle color shift across the face: rosier near the nose and cheekbones, more even and neutral toward the temples, real skin, not a uniform digital paint fill

Close portrait, [SUBJECT], visible natural blush at the cheeks and nose from surface blood vessels, contrasted with a cooler, more neutral forehead and jawline, photographic realism, no smoothing

Soft light: for a smoother, still-real look (3)

Portrait of [SUBJECT], large soft diffused light source (overcast sky or big softbox), wraps around the face and minimizes harsh shadows, skin still shows faint pore texture and a soft highlight rather than going flat or plastic

[SUBJECT] near a north-facing window with sheer curtain, soft even light, gentle shadow falloff, subtle skin texture retained, flattering but not artificially smoothed

Beauty-adjacent portrait, [SUBJECT], softbox lighting at 45 degrees, smooth but not poreless skin, one faint natural highlight on the forehead, avoiding an airbrushed or CGI finish

Hard light: for visibly revealed texture (3)

Portrait of [SUBJECT], single small hard light source (direct sun or a bare bulb) at a low raking angle, short-edged shadows revealing every pore and fine hair on the skin's surface, high-texture, documentary realism

[SUBJECT] lit by harsh midday sun from the side, defined shadow under the nose and cheekbones, visible skin texture and imperfections fully revealed by the hard light, not softened

Gritty, unpolished portrait of [SUBJECT], small hard key light close to the lens axis, texture-revealing shadows across the skin, no diffusion, no retouching, honest and unglamorous

Imperfections that read as real (3)

Portrait of [SUBJECT], a scatter of light natural freckles across the nose and upper cheeks, one faint asymmetry to the smile, otherwise clean skin with visible pore texture, no over-correction toward flawlessness

[SUBJECT], subtle fine lines at the outer corners of the eyes, natural uneven tone across the cheeks, real unretouched skin, avoiding both an airbrushed look and an exaggerated "aged" filter

Documentary-style portrait, [SUBJECT], faint acne scarring at the jawline, natural skin texture elsewhere, treated as a real physical detail rather than something to hide or dramatize

Skin in shadow: keeping it alive, not flat (3)

Portrait of [SUBJECT], dramatic side lighting, strong key light on one side of the face, soft fill light on the shadowed side preserving subtle skin texture and a faint natural glow rather than going flat black

[SUBJECT] in low evening light, gentle bounce or ambient fill keeping a hint of warmth and texture visible in the shadowed half of the face, avoiding pure matte darkness

Moody portrait, [SUBJECT], hard key light with a dim reflector fill, shadow side of the face still shows a trace of specular highlight and skin tone, not a flat silhouette

Makeup vs. bare skin: pick one target (3)

Portrait of [SUBJECT], bare skin, no makeup, visible natural pore texture, fine hair, and color variation, soft natural light, completely unretouched look

Portrait of [SUBJECT], full glam makeup, matte foundation evening the tone, contoured cheekbones and soft-focus finish, polished editorial beauty look, intentional smoothness as a makeup effect, not a rendering flaw

[SUBJECT], minimal "no-makeup makeup" look, slight evening of tone at the cheeks only, pore texture and fine hair still visible everywhere else, natural finish

Prompting a specific skin tone, respectfully (4)

Portrait of [SUBJECT], deep brown skin with warm golden undertones, soft key light metered and exposed for this skin tone, visible pore texture and a natural highlight at the forehead and cheekbones, shadow side retains subtle detail and glow

Portrait of [SUBJECT], fair skin with cool pink undertones, soft daylight, visible pore texture and fine hair, careful exposure to avoid blowing out highlight detail on lighter skin

Portrait of [SUBJECT], olive skin with neutral to warm undertones, natural outdoor light, visible texture and subtle color variation at the nose and cheeks, exposure balanced for this specific tone rather than a generic default

Portrait of [SUBJECT], tan skin with warm undertones, directional light revealing pore texture and fine hair, shadow side of the face lit with a soft fill so detail and warmth are not lost

Fixing skin that already looks plastic (2)

Append to any existing prompt: add visible pore texture, subtle color variation across the face, and a soft natural highlight instead of an even matte or glossy finish; remove any "smooth," "flawless," or "airbrushed" language already in the prompt

Revise for realism: replace "perfect skin" or "flawless complexion" with "unretouched skin with visible pores, fine lines, and natural color variation"; replace "studio beauty lighting" with a named directional light source

Negative-prompt style exclusions, where the tool supports them (2)

avoid: airbrushed skin, poreless skin, waxy or plastic texture, uniform flat skin tone, glossy all-over shine, CGI or 3D-render skin appearance

avoid: overly smoothed skin, mannequin-like texture, uncanny symmetrical perfection, painted or matte-plastic finish

Support for a literal negative-prompt field varies by tool, and stacking words like these into the main prompt as "avoid" language works more consistently across models than assuming a dedicated field exists everywhere. See the negative prompt support matrix for which tools actually have one.

Two notes on the placeholders. [SUBJECT] always resolves to a described, invented person, never a named real one, per the rule above. [LIGHT SOURCE] and [SETTING] are yours to swap for whatever the shot needs; the surrounding skin vocabulary is doing the real work and travels across scenes. Pair these with camera and lens terms that change your AI images if lens choice matters as much as light, or with free AI headshot prompts for the framing and aspect-ratio side this post doesn't cover.

On tools: Black Forest Labs pitches FLUX.2 specifically at this problem, describing "4MP photorealistic output" that's "closing the gap with real photography" on detail (bfl.ai/models/flux-2, accessed September 2026). That's the vendor's own framing, not an independent ranking, worth reading as a place to start testing rather than a verdict.