50 Tests Later: The Frost-Covered Goggles Prompt That Worked

Extreme macro close-up of frost-encrusted matte black ski goggles worn by an alpine mountaineer, the lens surface treated as the primary subject. ::2.4 The mirrored lens carries a compressed, low-contrast, slightly desaturated reflection of a windswept snow-covered ridgeline held at roughly 40 percent of the frame's luminance, with a small silhouetted climber rendered as a dark shape rather than a detailed figure. ::1.4 Intricate dendritic frost crystals branch along the goggle frame edges only, with individual frozen condensation droplets on the lens surface exhibiting subsurface scattering and refraction, index of refraction near 1.31, catching micro-highlights of one to two pixels at 8K. ::1.1 A piercing low-angle winter sun produces a narrow diagonal lens flare, its core a warm amber (2800K) fringed by distinct cyan (7500K) chromatic aberration along high-contrast edges only, never across the frost. ::0.9 Weather-beaten skin visible at the nose bridge and upper cheek, wind-burned red with fine ice particles, textured against the coarse dark fibers of a wool balaclava whose knit scale sits measurably below the pore scale of the skin. ::0.8 Background conveys thin crisp mountain air with subtle backlit falling snow particles catching golden rim light, held in creamy bokeh to preserve atmospheric depth. :: Photorealistic, hyper-detailed 8K, emphasizing tactile ice texture, translucency and frozen condensation micro-detail. :: Shallow depth of field isolates the frost formations and skin texture. :: Cinematic color science: crushed blacks set at 5 IRE, shadows lifted with a subtle cyan tint, highlights rolling off warm at 3200K, rendered on a log gamma curve. :: Shot on an Arri Alexa 65 with a Master Macro 100mm T2.8 wide open at f/2.8, 1.4x macro magnification, 4:3 sensor extraction, vertically framed. --ar 9:16 --style raw --s 250 --q 2

Set the lens weight too high and the goggles stop being goggles. The mirrored surface renders as an opaque chrome plate sitting at 85 to 95 percent luminance, the ridgeline reflection flattens into a single gradient with no separation between snow and sky, and the climber silhouette becomes an indistinct grey smudge. Worse, the dendritic frost that should sit on the frame edge dissolves into clipped white pixels that are numerically identical to the specular blowout. The nose bridge reads as wax. The defect is not the snow or the frost — it is the missing tonal hierarchy between them.

The opposite failure is quieter and harder to diagnose. When frost and ice-detail clauses are weighted upward, the crystals render as high-frequency noise, the reflection drops below 25 percent luminance, and it merges with the crushed blacks set at 5 IRE. The ridgeline disappears into shadow, the climber vanishes entirely, and the lens reads as a dirty window rather than a reflective surface. Both failures share one cause: the prompt never states where the reflection sits in the tonal range, so the model assigns it maximum contrast and maximum detail by default.

Generic quality tags make this worse. Phrases like "photorealistic, hyper-detailed 8K" push every surface toward maximum micro-contrast simultaneously, which flattens the hierarchy instead of enriching it. A macro frame only reads as macro when some planes are deliberately softer, dimmer or less detailed than others. The repair is not to remove those tags but to subordinate them with explicit, measurable constraints on the one element that governs the whole image: the mirrored lens.

Weights Are Ratios, Not Intensities

Midjourney's multi-prompt syntax uses double colons to separate concepts and assign relative weights, and those numbers behave as ratios rather than absolute intensities. A clause at ::2.4 competing with a clause at ::1.4 is a 1.7-to-1 relationship, not a brightness value. The practical consequence is that adding any new weighted clause silently re-balances everything already present. Insert a ::1.6 falling-snow clause and the lens reflection loses effective share even though its own number never changed.

This is why the fix has to be structural. If you tune the reflection by editing its number alone, you will chase it indefinitely as other clauses move. Instead, lock the relationship: the lens is the subject at the highest weight, frost is a competing texture layer at roughly 60 percent of that weight, skin and balaclava sit below frost, and background snow sits lowest. Every later edit must preserve those proportions. You can verify the ratio arithmetic against the official Midjourney documentation before committing to a long render queue.

The 2.4 weight does one specific job: it declares the lens as the primary subject, which prevents the model from treating the goggles as a prop on a face. The 1.4 frost weight does a different job — it keeps frost as a visible competitor that loses. If frost were weighted at 0.8, the frame edges would render nearly clean and the image would lose its entire reason for existing. The gap between 2.4 and 1.4 is the article's central mechanism, and it is the first thing to protect when iterating.

The 40 Percent Luminance Budget

The reflection is held at roughly 40 percent of the frame's luminance, and that number is not decorative. Above 55 percent, the reflection and the frost highlights clip into the same white, and the viewer's eye can no longer separate a distant ridgeline from a near crystal. Below 30 percent, the reflection collapses into the shadows lifted with a cyan tint and the ridgeline loses its silhouette. Forty percent sits in the usable middle where both a bright snowfield and a dark climber shape can coexist inside one small reflective area.

There is a second-order interaction worth accounting for. The prompt specifies highlights rolling off warm at 3200K on a log gamma curve, and a warm rolloff compresses the top end rather than clipping it. A reflection authored at 40 percent can therefore land at 48 to 52 percent after tone mapping. That headroom is useful, but it means you should not author the reflection at 45 percent and expect it to stay there. Authoring low and letting the rolloff lift it is the safer direction.

Low contrast and slight desaturation inside the reflection are depth cues, not stylistic preferences. A distant snowfield photographed through a mirrored lens genuinely carries less contrast and less saturation than objects at the frame plane. Specifying both properties tells the model that the reflection belongs to a different depth layer than the frost. This is the same layering logic used in hyper-realistic collage composition, where depth is communicated through relative contrast rather than through blur alone.

The crushed blacks at 5 IRE matter here too. Setting them to absolute zero removes the shadow detail that keeps the climber silhouette readable as a shape. Five IRE is dark enough to read as black in a cinematic frame while retaining just enough separation for a human form to register. The climber should be described as a dark shape rather than a detailed figure precisely because at that luminance there is no room for detail — and asking for detail would force the model to brighten the silhouette, breaking the reflection's budget.

Refractive Index and Why "Frosty" Fails

"Frosty" is a mood word. A diffusion model receiving it must invent the physics, and the cheapest invention is metallic specular reflection, which is exactly the chrome-plate failure described earlier. Stating an index of refraction near 1.31 — the value for water ice in the visible spectrum — pushes the material toward dielectric behavior instead. Dielectrics reflect a small fraction of light at normal incidence and transmit the rest, which is what makes a frosted lens look like ice rather than foil.

Subsurface scattering and refraction are listed together for the condensation droplets, and the pairing is deliberate. A droplet a few millimeters across is dominated by surface refraction, but its edges scatter light into a soft rim. Requesting both, anchored by the refractive index, produces a consistent material rather than two competing ones. The trade-off is real: raise subsurface scattering too far and the droplets turn milky and opaque, losing the crisp micro-highlights that make them read as frozen rather than fogged.

Specifying micro-highlights of one to two pixels at 8K is a scale instruction disguised as a detail instruction. Left unspecified, a model renders "sparkle" as a broad soft glow covering dozens of pixels, which reads as bloom rather than ice. Two pixels is small enough that the highlights behave like point sources. Note the resolution dependency: at a native output width of 1024 pixels, a two-pixel highlight is roughly 0.2 percent of the frame width, so the effect only survives if you upscale deliberately rather than relying on the base render.

Spatial localization is the other half of this clause. Dendritic frost branches along the goggle frame edges, not across the lens center. Frost grows where moisture collects and where the frame is coldest, which in a real goggle means the rim and the vent edges. Telling the model where frost is allowed to appear prevents it from scattering crystals across the reflection, which would destroy the 40 percent luminance budget in a single pass. The same material-specific instruction pattern appears in ultra-realistic portrait work, where fabric and skin must not share a texture frequency.

Edge-Localized Aberration and Scale Hierarchy

Chromatic aberration is specified along high-contrast edges only, and that restriction is the difference between optics and noise. Real lateral CA scales with local contrast and distance from the optical axis, so it appears on the sun flare's boundary and on the hard silhouette of the climber, not on every frost dendrite. Applied globally, a 2800K amber core fringed with 7500K cyan turns each crystal into a rainbow artifact, and the frame reads as a compression error rather than a lens rendering.

The temperature pairing itself is a signature rather than a grade. A warm amber core with a cool cyan fringe is the classic look of lateral chromatic aberration on a low-angle winter sun, because the two ends of the spectrum focus at slightly different planes. Writing it as a temperature pair gives the model a physically coherent artifact. Writing it as "color fringing" gives it permission to place rainbow edges anywhere, including across the skin, which instantly breaks the realism of the nose bridge.

Scale hierarchy is stated explicitly: the wool knit sits below the pore scale of the skin. This ordering is what prevents the balaclava from reading as skin or the skin from reading as fabric. When two materials render at the same spatial frequency, the eye cannot assign them different distances from the lens, and the macro illusion collapses. Naming the ordering — knit finer than pores, ice particles finer than knit — gives the model a three-level frequency ladder that matches the depth of field.

Wind-burned red skin with fine ice particles must stay below the frost in detail density, or the face competes with the lens for attention. The red is a hue instruction, not a saturation instruction; pushed too far it turns orange under a 3200K highlight rolloff. Keeping skin as the third-highest weighted element, behind lens and frost, is what keeps the composition anchored on the goggles while still giving the frame a human scale reference.

Optical Chain and the Acceptance Checklist

A 100mm master macro at f/2.