Extreme close-up portrait, woman's right eye and cheekbone, hyperrealistic digital art. Golden-amber iris with visible collarette fibers and limbal ring. Rainbow gradient eyeshadow: cobalt blue at inner corner, transitioning through emerald, chartreuse, tangerine, to crimson at outer brow. Individual black lash hairs with micro-highlighting. Vertical paint drips: saturated pigment streams in crimson, tangerine, canary, lime, cobalt, violet, melting from temple down cheek with viscous liquid physics — thicker accumulation at origin points, thinning into tendrils. Water droplets: 20-30 scattered across skin, each with caustic light refraction and meniscus distortion. Single tear: crystalline, tracking from outer canthus with surface tension bead at terminus. Glossy crimson lips, slightly parted showing upper teeth edge. Lighting: 5500K key light 45° camera left through 4x4 diffusion, 3200K rim light 90° camera right for warm edge separation, fill at -2 stops. Photorealistic skin: visible pores on cheek, fine vellus hairs, subtle sebum sheen on nose bridge. Shallow depth of field: iris tack-sharp, earlobe soft. 8K, editorial beauty photography, emotional surrealism --ar 9:16 --style raw --s 750
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 and customize to make it your own!
The prompt does not read as a sentence. It reads as a sequence of weighted token groups, and the model commits to a composition decision within the first handful of them. "Extreme close-up portrait, woman's right eye and cheekbone" is doing more work than any adjective that follows, because it locks the crop before the style tokens arrive. Change that phrase to "portrait of a woman" and every later instruction - iris fibers, lash micro-highlighting, drip viscosity - gets rendered at a fraction of its intended resolution. The machine allocates pixel budget based on the earliest spatial claim it can parse. This is why the framing clause must precede any material or lighting descriptor.
Once the crop is fixed, the model begins resolving what it treats as the subject's defining surface. Here that surface is the iris, not the face. "Golden-amber iris with visible collarette fibers and limbal ring" gives the renderer three separate anatomical constraints, and each one forces a different level of magnification. Collarette fibers only become visible at macro distance. Limbal ring only reads at macro distance. The model infers that the entire image must sustain that magnification, which is what keeps the eyelashes sharp and the earlobe soft rather than the reverse. Specificity of a single feature dictates global depth of field.
Decomposing the Rainbow Eyeshadow Gradient
The eyeshadow instruction is a color ramp, not a color list. "Cobalt blue at inner corner, transitioning through emerald, chartreuse, tangerine, to crimson at outer brow" specifies both endpoints and the path between them. A prompt that says "rainbow eyeshadow" alone produces banded, arbitrary hues because the model has no anchor points. Naming the terminal colors and the intermediate stops forces interpolation. This is the same logic that governs any gradient work, including the layered pigment approach described in this impasto night scene breakdown, where color transitions are pinned to physical paint behavior rather than left to the sampler.
Directionality matters as much as the hues. Specifying "inner corner" and "outer brow" as the two anchors tells the renderer the gradient runs along the eyelid's diagonal axis, not radially from the pupil. Radial gradients read as cosmetic application. Diagonal gradients read as painted or dripped pigment. The distinction determines whether the image looks like beauty editorial or body art, and it costs four words to control.
The lash clause, "individual black lash hairs with micro-highlighting," is a resolution demand disguised as a texture note. "Individual" tells the model that lashes cannot be merged into a mascara mass. "Micro-highlighting" tells it that each hair must carry its own specular response. Together they push the sampler toward per-strand rendering, which is only achievable when the depth-of-field token later confirms macro focus. Remove either word and the lashes collapse into a single dark shape.
Why Drip Physics and Water Droplets Must Be Separated
Vertical paint drips and water droplets are physically different phenomena and the prompt treats them as such. The drip clause specifies viscosity: "thicker accumulation at origin points, thinning into tendrils." That is a description of fluid behavior under gravity, and the model renders it as a continuous mass with varying width. The droplet clause specifies count and optics: "20-30 scattered across skin, each with caustic light refraction and meniscus distortion." That is a description of discrete spheres with surface tension. Merging these into a single "wet colorful face" instruction produces neither. The renderer averages the two behaviors and outputs a uniform glossy sheen.
The droplet count is not decorative. Stating a number between 20 and 30 prevents the model from either flooding the skin with moisture or omitting it entirely. Without a count, droplet density follows the sampler's default tendency toward even distribution, which reads as artificial. A range forces irregular clustering, and irregular clustering is what makes the skin read as skin rather than as a rendered surface. The meniscus distortion token is what sells the refraction at the droplet edge, where light bends most visibly.
The single tear is the only element in the prompt specified as unique. "Crystalline, tracking from outer canthus with surface tension bead at terminus" gives it a path, a material, and a terminal state. The surface tension bead is the critical token. Without it, tears render as flat streaks. With it, the renderer understands that the liquid terminates in a rounded volume held by its own cohesion, which is the visual cue that separates a photographed tear from a painted one. Compare this to the control exercised in dramatic feathered portrait prompts, where single-element specificity similarly carries the emotional read of the frame.
Lighting Ratios, Skin Tokens, and the Editorial Finish
The three-light setup is specified in Kelvin and degrees, which is unusual and effective. "5500K key light 45° camera left through 4x4 diffusion" tells the renderer both the color temperature and the softness of the source. Diffusion size relative to subject determines shadow edge quality, and naming a 4x4 frame communicates a soft but directional falloff. The 3200K rim at 90° camera right creates a warm separation edge against the background, and the -2 stop fill prevents the shadow side from crushing to black. This is a standard editorial ratio expressed in machine-readable terms.
Skin realism is handled through three tokens that operate on different scales: pores, vellus hairs, sebum sheen. Pores are micro-texture, vellus hairs are mid-frequency detail, sebum sheen is a specular property. Each one is rendered by a different part of the pipeline, and each one fails differently when omitted. Without pores the skin looks plastic. Without vellus hairs the cheek edge looks cut out. Without sebum sheen the nose bridge looks matte and dead. Naming all three is what prevents the model from defaulting to its smoothed beauty-retouch prior.
The closing tokens, "8K, editorial beauty photography, emotional surrealism," function as a style prior that biases the sampler toward magazine-grade finishing. Where a cleaner commercial look is needed, generators like Midjourney respond to these style anchors more predictably than to longer descriptive chains, particularly at the raw setting where the model resists over-stylization. The --ar 9:16 ratio is not cosmetic either: a vertical crop forces the composition to prioritize the eye-to-lip vertical axis, which is exactly the axis the drip and tear instructions run along.
The boundary of this approach appears when the prompt tries to carry more than one subject. The entire structure depends on a single eye owning the frame. Add a second face, a hand, or a background environment and the token budget spreads, the macro focus claim collapses, and the iris fibers, lash strands, and droplet counts all degrade simultaneously. The signal that the limit has been crossed is simple: if the iris is no longer the sharpest element in the render, the prompt has exceeded its scope and needs to be split into two images.
Label: Fashion
Key Principle: Separate pigment layers from liquid layers in the prompt: specify shadow gradients first, then drip physics second. The model assigns separate render passes to each, and merging them collapses both into flat color.
Comments
Post a Comment