Eco-Luxury Skincare Prompts: Fixing Shallow Depth

A hyper-realistic commercial product photograph of a 30ml cylindrical serum dropper bottle, pale aquamarine translucent borosilicate glass with a measured internal gradient running from saturated aquamarine at the base to fully clear at the shoulder, mirror-polished chrome silver collar carrying one continuous softbox edge highlight, matte white nitrile rubber bulb with fine molded surface texture. Bottle centered at eye level on weathered bleached driftwood with deep grain fissures, entwined with dense emerald-green sphagnum moss and three fresh young leaves with visible venation. Single 5600K softbox key from upper left at 45 degrees, 24x36 inch diffusion panel at 2 feet distance, producing luminous caustic refraction through the glass onto the moss surface, a specular metallic highlight band on the chrome collar, and soft-edged dappled shadows at a 3:1 falloff ratio across moss textures. Clean atmospheric powder blue seamless background at 6 feet distance with subtle vertical gradient falloff to white, preserving 40% negative space above the product for text overlay. Color palette balances cool aquamarine and seafoam tones against warm sandy driftwood browns and vibrant chlorophyll greens. Shot at f/8 for uniform sharpness, 100mm macro lens, medium format digital sensor, editorial beauty photography aesthetic with controlled film grain. Botanical elements frame the product without obstruction, driftwood grain and moss fiber detail rendered at macro clarity, product base contact shadow anchoring it to the surface. --ar 2:3 --style raw --v 6

A naive eco-luxury skincare prompt reads like a shopping list: "luxury serum bottle on moss, soft light, blue background, beautiful." The render it returns shows glass holding one flat value of aquamarine, a white stripe where a chrome collar should be reflecting something, and moss that behaves like green carpet. An informed prompt for the identical subject specifies a 30ml cylindrical dropper in translucent borosilicate, an internal gradient from saturated base to clear shoulder, and a 5600K softbox at 45 degrees through a 24x36 inch diffusion panel two feet away. These are not variations of one another. One is a drawing of a bottle; the other is a photograph of light passing through a bottle.

The gap between them is not vocabulary richness. It is whether the prompt encodes optical events a renderer can actually simulate. Diffusion models learn statistical correlations between text tokens and pixel patterns, so "beautiful" correlates with high-key contrast and shallow depth of field, not with a specific caustic landing on moss. Physics terms — transmission, refraction, specular band, falloff ratio — correlate with the pixel structures those phenomena produce. Adjective density changes the mood of an image. Physical specificity changes what is visible inside it.

The Problem: Uniform Glass, Painted Highlights, Dead Moss

The first failure is value flattening in transmissive material. A prompt that says "blue glass bottle" without naming a gradient direction gives the model no reason to vary luminance across the vessel. The result is a single mid-tone aquamarine filling the silhouette from base to shoulder, reading as tinted plastic rather than 3mm borosilicate. Glass reads as glass because thickness changes how much light escapes, and thickness varies along a cylinder. If the prompt never states that the base is saturated and the shoulder is clear, no gradient is generated.

The second failure is the highlight as painted stripe. Chrome collars in naive renders show a uniform white band with hard top and bottom edges, which is what a graphic designer draws, not what a mirror-polished cylinder does. A real collar reflects the shape of the light source: a rectangular softbox produces a rectangular highlight with soft falloff at its edges, and its position on the collar is determined by the camera-to-light angle. Without an angle, the model defaults to a centered band.

The third failure is botanical deadness. Moss described only as "green moss" renders as a low-frequency texture map with no fiber structure and no interaction with the product. Two things are missing: the contact shadow where the bottle base meets the driftwood, and the caustic — the bright refracted patch that transmissive glass throws onto whatever sits beneath it. Both are consequences of light geometry, and both are absent when the prompt describes objects instead of illumination.

Root Cause: Adjectives Occupy Tokens That Physics Needs

Every prompt has a finite attention budget, and tokens spent on subjective qualifiers compete with tokens spent on measurable parameters. "Luxurious," "premium," "stunning," and "high-end" pull attention toward global contrast and saturation curves. They do not pull toward the collar's reflection geometry or the gradient inside the glass. In practice, a prompt with six luxury adjectives and no light distance produces a more saturated image than a prompt with zero adjectives and a stated 2-foot diffusion distance — and a less believable one.

Order compounds the problem. Tokens appearing earlier in a prompt tend to carry more weight in the conditioning signal, so a subject description buried under mood language arrives weakened. When material properties such as "translucent borosilicate" and "mirror-polished chrome" are stated before atmospheric language, the model resolves the object's surface behavior first and applies the background gradient afterward. Reversed, the background mood bleeds into the object and the glass picks up a hazy, low-contrast look.

Material conflicts are the subtlest cause. "Glossy" and "matte white nitrile" in the same sentence, or "soft light" combined with "crisp specular highlight," give the model contradictory targets and it averages them. The averaging is visible: a collar that is neither mirror nor satin, a bulb that is neither rubber nor porcelain. Naming each material once, with its finish, removes the ambiguity that produces these hybrid surfaces.

The Fix: Rebuild the Prompt as a Light and Material Specification

Start with the material stack, ordered from the outside in. Vessel: 30ml cylindrical dropper, translucent borosilicate, internal gradient saturated at base to clear at shoulder. Closure: mirror-polished chrome collar. Applicator: matte white nitrile bulb with fine molded texture. Naming the finish of each component separately is what prevents the renderer from blending them into one averaged surface, and it is the same discipline used in luxury material rendering for metallic and reflective surfaces, where finish language is separated per element rather than applied to the whole scene.

Then specify the light as an instrument, not a mood. One 5600K softbox key, upper left, 45 degrees, 24x36 inch diffusion panel at 2 feet. Each number removes a degree of freedom: the color temperature fixes the white balance of the highlights, the angle fixes where the collar's reflection band lands, the panel size fixes how soft the highlight's edges are, and the distance fixes the falloff ratio. A 3:1 ratio across the moss is a measurable target — the lit side of a moss clump should sit roughly one and a half stops above the shadow side.

Finally, name the optical results you expect, because naming them reinforces the parameters that produce them. Caustic refraction through the glass onto the moss. Specular metallic highlight band on the collar. Soft-edged dappled shadows. Product base contact shadow. These phrases are not decoration; they are the visible consequences of the geometry already specified, and including them tightens the correlation between your parameters and the pixels. Layering surface detail this way follows the same logic as building hyper-realistic texture through stacked material cues rather than single descriptive words.

Composition and ratio close the specification. Centering the bottle at eye level, placing the background at 6 feet with a vertical gradient to white, and reserving 40% negative space above the product all serve one purpose: keeping the frame usable for text overlay. The 2:3 aspect ratio at --ar 2:3 gives a vertical canvas where that negative space sits above the subject rather than beside it. --style raw reduces the model's default aesthetic post-processing, which otherwise lifts contrast and desaturates shadows in a way that erases the 3:1 falloff you specified.

Verifying the Fix in the Render

Check the base first. A correct render shows a bright, irregular caustic patch on the moss directly beneath and slightly forward of the bottle, its shape following the curvature of the glass base. Its absence means the renderer treated the glass as opaque or the light distance was too large to concentrate transmission. Parameter documentation for flags such as --style raw and --ar is maintained at Midjourney's official site, and it is worth confirming there that your version supports the flags you are relying on before blaming the prompt.

Check the gradient next, at 100% zoom. Luminance should rise smoothly from the saturated base to the clear shoulder over the full height of the vessel, with the transition occupying roughly the lower two-thirds. A hard horizontal line means the model interpreted the gradient as a graphic element rather than a thickness effect. A completely uniform fill means the gradient clause was truncated — a common outcome when material language sits past the point where the conditioning signal thins out.

Check the collar and the shadow together, since they share the same light geometry. The specular band should sit off-center, on the side facing the 45-degree key, with soft horizontal edges matching the diffusion panel's proportions. The contact shadow under the bottle should be tight and dark at the base, opening into the broader dappled moss shadow. If the collar highlight is centered and the shadow is absent, the light clause was overridden by earlier scene language — reorder and re-render. Adobe's Firefly documentation describes comparable structure-versus-style controls for its generative models, useful when porting the same specification across platforms.

Where This Approach Stops Working

This method depends on transmissive and reflective materials. When the packaging is opaque matte — a kraft carton, a stoneware jar, a soft-touch tube — there is no internal gradient to specify and no caustic to generate, so the light-geometry clauses stop producing visible returns. The signal that you have crossed the boundary is prompt growth without render change: you add a fourth optical term and the image is pixel-identical, because there is no transmissive surface for it to act on.

A second boundary is token dilution. Past a certain prompt length, later clauses lose influence, and the gradient direction or falloff ratio you placed at the end simply stops registering. The diagnostic is a render that satisfies the first half of your specification and ignores the second. The response is not more words but fewer objects: cut the botanical count, drop the secondary reflections, and keep the parameters that produce visible events.

Read the next render as a measurement, not a verdict. Compare the caustic's position against the stated 45-degree angle, compare the shadow side against the 3:1 target, and compare the gradient's transition height against the lower two-thirds expectation. Each mismatch points to one clause, and one clause is always cheaper to fix than a rewrite.