The AI Portraiture Approach That Clicked
Quick Tip: Click the prompt box above to select it, then press Ctrl+C (Cmd+C on Mac) to copy. Paste directly into Midjourney, DALL-E, or Stable Diffusion!
The Physics of Convincing Light
The difference between a passable AI portrait and one that arrests attention often comes down to how light is described. Most prompts fail because they treat light as an atmosphere rather than a physical phenomenon. "Dramatic lighting," "golden hour," "soft and ethereal"—these phrases trigger the model's aesthetic averaging, producing results that feel familiar because they are, literally, average of millions of training examples.
The breakthrough comes from understanding that light in photography is not decoration. It is information about the world. A hard shadow edge tells us the light source is small and distant. A warm bounce in the shadows reveals a nearby colored surface. The specific pattern of venetian blind shadows communicates not just mood but time of day, window orientation, and room geometry. When you describe these physical properties rather than their emotional effect, you give the AI concrete parameters to simulate.
Consider how shadow works in actual photography. A venetian blind creates parallel shadow bands because the slats are parallel and the light source (the sun) is effectively a point source at distance. The shadow sharpness depends on the angular size of the source—morning sun is lower, so shadows stretch longer but remain sharp because the sun's apparent size doesn't change. Describe this mechanism—"hard morning sunlight through venetian blinds"—and the AI generates shadows with correct geometry: parallel, sharp-edged, with spacing that suggests slat width and sun angle. Describe "striped shadows for mood" and you might get any pattern that feels striped, regardless of physical plausibility.
The color temperature specification serves a similar function. Human vision adapts to color temperature constantly; we perceive white paper as white under tungsten or daylight. Cameras do not—they record the actual spectral content. A 5600K daylight source through glass, bouncing off cream walls and warm wood floors, creates a complex color environment that reads as natural because it follows physical rules. The key light cools the illuminated skin toward neutral-warm; the bounce fill warms the shadows. This separation of color temperatures across the tonal range creates dimension that uniform "warmth" cannot achieve.
Skin as Material, Not Style Category
Perhaps no element of AI portraiture fails more consistently than skin rendering. The problem lies in how we prompt it. "Realistic skin" or "beautiful skin" or "natural skin" all trigger the model's conception of skin as a quality to be optimized—smoothed, even-toned, perfected. Actual human skin is none of these things under close inspection. It is porous, slightly oily, with microscopic hairs and transient color variations from blood flow near the surface.
The solution is to treat skin as a material with specific optical properties under specific lighting conditions. Pores become visible when light grazes the surface at shallow angles; this is why beauty photography often uses large soft sources or ring lights to fill them. Peach fuzz catches rim light, creating a subtle halo that separates the subject from background. Sebum produces small, bright specular highlights that signal healthy skin rather than plastic smoothness.
When you specify these properties under stress—"pore detail retained in illuminated areas," "subtle peach fuzz catching edge light"—you prevent the model from applying its default smoothing. The key is describing skin in its interaction with light, not as an isolated quality. The light reveals the material; the material validates the light. This reciprocity is what makes rendered skin feel present and embodied rather than generated.
The shadow areas demand equal attention. AI models tend toward high contrast as a shortcut to "dramatic," crushing shadow detail to pure black. Professional portraiture maintains detail throughout the tonal range—shadows are dark but not absent. Specifying "shadow areas hold detail without crushing" forces the model to manage its dynamic range more carefully, preserving the subtle information (fabric texture, background separation, secondary light sources) that gives the image depth.
Lens Physics and Focal Choices
The 85mm lens has become a cliché in portrait photography, but its prevalence reflects genuine optical properties. At this focal length on a full-frame sensor, facial features retain their natural proportions without the widening distortion of wider angles or the flattening compression of telephoto extremes. The subject-to-background distance that produces pleasing framing also naturally throws backgrounds out of focus.
Specifying aperture—f/1.4 in this case—adds critical information about depth of field behavior. The AI understands that only a thin plane will be sharp, and it typically places this plane on the eyes or the closest eye in a turned head. Without aperture specification, the model defaults to a middle-ground that often leaves too much sharp, producing the "everything in focus" look of computational photography or small-sensor cameras.
The combination of focal length and aperture also affects bokeh quality—the character of out-of-focus areas. An 85mm f/1.4 produces smooth, circular bokeh with slight cat-eye distortion at frame edges. This is distinct from the busier bokeh of telephoto zooms or the geometric patterns of mirror lenses. While the AI may not render these subtleties perfectly, the specification guides it away from the uniform Gaussian blur that signals "artificial background separation."
Color Temperature as Environmental Information
Color temperature in photography is not merely white balance correction—it is data about light sources. Daylight is approximately 5600K. Tungsten bulbs are roughly 3200K. Overcast sky can reach 7000K or higher. When multiple sources with different temperatures illuminate a scene, the camera records this as color contrast that our visual system would normally neutralize.
In the venetian blind portrait, the primary source is 5600K daylight—cool, blue-white, direct. But light also bounces off surfaces: the cream walls, the wooden floor, the white sweater itself. These bounces pick up the color of their reflecting surface. A wooden floor at 3200K (warm amber) becomes a secondary fill source, warming the shadow side of the face that receives no direct daylight. This creates the characteristic look of interior daylight portraiture: cool key, warm fill, dimensional color.
Without explicit temperature specification, the AI tends toward uniform warmth or coolness based on mood keywords. "Morning light" might produce warm yellows throughout; "soft light" might desaturate entirely. The specific Kelvin values anchor the model in physical behavior, producing the complex color interactions that signal "photographed" rather than "generated."
From Prompt to Practice
The principles here extend beyond this single image. Any portrait prompt benefits from physical specificity: where is the light coming from, what is its quality, what does it pass through or reflect from, how does it interact with specific materials? The venetian blind is one occlusion pattern; window sheers, tree canopy, partially closed curtains, or architectural elements like colonnades create others. Each produces a distinctive shadow geometry that communicates specific environmental information.
The material descriptions—chunky knit with visible weave, off-shoulder construction showing collarbone and shoulder, pushed-up sleeves creating volume at the forearms—give the AI information about how fabric falls, stretches, and catches light. These details matter because they create the micro-variations in texture and shadow that signal physical presence. A sweater described only as "white" becomes flat; a sweater described as "chunky knit with visible cable pattern, off-shoulder drape showing skin contact at neckline" becomes an object with weight and behavior.
The final element is restraint in stylization. The --style raw parameter in Midjourney disables the model's default aesthetic processing, which tends toward pleasing but generic results. Combined with --s 250—a middle-low stylization value—it allows the prompt's specific instructions to dominate without drifting into the uncanny smoothness of high-stylization portraits. The result retains the slight imperfections and specificities that mark human photography: pores, fabric texture, asymmetric shadow falloff, the particular quality of light through actual blinds.
This approach—physical specification over mood, material interaction over quality adjectives, optical parameters over generic camera references—transforms AI portraiture from lottery to craft. The image becomes predictable not because it is repetitive, but because the underlying physics are consistent. Light behaves as light should. Materials respond as materials do. The result is not just an image that looks like a photograph, but an image that behaves like one.
For related approaches to controlled lighting in different contexts, see mastering dramatic feathered portraits and mastering Midjourney street portraits. The underlying principles of light quality and material specificity apply across genres, from studio fashion to environmental portraiture.
Official Midjourney documentation and community resources are available at midjourney.com.
Label: Fashion
Key Principle: Replace mood words with physical specifications: every light needs a source, every shadow needs an origin, every material needs surface properties under stress.