TL;DR: Metal, glass, fabric and skin all bounce light differently, and "more detail" doesn't fix a render that has the wrong kind of reflection. Metal tints its highlight; glass lets light through and bends it; fabric and skin mostly scatter it. Below: how to name that difference, plus what vendors actually document versus what's borrowed studio vocabulary.
One thing before any of this: Prompt Architects writes the prompt, not the image. Nothing here renders a picture; these are words to paste into whichever model does.
Why Do Metal, Glass, Fabric and Skin Need Different Prompts?
Because each one handles the same light differently, and a diffusion model has to guess which behavior you mean unless you say so. Four behaviors cover almost everything you'll ever prompt: how much light bounces straight back at the viewer (specular reflection), whether that bounced light keeps the surface's own color or stays neutral, whether light passes through the material at all (transmission), and whether it enters and scatters just beneath the surface before leaving somewhere else nearby.
Metal reflects nearly all of it, tinted by the metal's own color. Glass reflects a little and lets the rest through, bent by refraction. Fabric and most solids scatter incoming light diffusely, with little to no coherent reflection. Skin does that same diffuse scattering, plus a small amount of light entering and re-emerging through its top layers. "Realistic texture" as a single instruction doesn't say which of these four behaviors the surface in front of you actually has, which is why it does almost nothing on its own.
How a material responds to light is a different question from what that light looks like in the first place; for the vocabulary describing the light itself, see lighting vocabulary for AI image prompts, which this post leaves entirely to that page.
What Makes an AI Render Default to Plastic?
When a prompt names an object's material weakly, "a metal cup," "a glass vase", without describing how light should behave on it, the output tends to land on a narrow, safe middle ground: a semi-glossy surface, a soft colorless highlight, no real transmission or tint. That's visually what plastic looks like, and it isn't a documented vendor behavior so much as the predictable result of an underspecified instruction landing on the most common "shiny object" pattern available.
OpenAI's own cookbook prompting guide states the fix as a general principle, not a materials-specific one: "Be concrete about materials, shapes, textures, and the visual medium (photo, watercolor, 3D render)" (developers.openai.com, GPT Image Generation Models Prompting Guide, accessed September 2026). The same guide's product-photography worked example asks for "realistic plastic and painted metal textures" side by side in one packaging shot, proof the model can render both correctly once each is named on its own terms rather than left for the model to guess between them.
Three things pull a render away from that default: naming whether the reflection carries the material's own color, naming whether light passes through it, and naming the reflection's sharpness. All three fail silently when a prompt only names the object.
How Do You Prompt Polished and Brushed Metal?
Two decisions come before any specific metal: how sharp the reflection is, and whether it carries the metal's own tint or stays neutral. Chrome, mirror-polish and clean steel reflect nearly the entire environment back in sharp, coherent detail, reading as colorless or faintly cool unless something colored sits directly in front of them.
Brushed or hairline-finished metal is different in a specific, describable way: fine parallel scratches from the finishing process scatter reflections along one axis only, stretching a highlight into a streak running parallel to the brush marks rather than sitting as a clean point or mirror image. That directional smear is what separates "brushed" from "polished" in a render, and naming the brush direction is what actually asks for it.
Gold, copper and brass reflect with the metal's own color baked into the highlight, not just the base surface: a gold object's brightest highlight is warm and yellow, not white, which is the detail most renders miss when a prompt says "gold" but describes the highlight as if the metal were chrome. Oxidation and patina sit on top of all this as a separate layer: verdigris on copper, rust on iron, tarnish on silver are surface coatings, and naming where they concentrate, edges, recessed detail, weather-exposed areas, reads as history rather than a uniform filter. If the object in frame is something you'd actually sell, a free product photography prompt generator is a reasonable starting point for the surrounding shot.
Close-up of [OBJECT], polished chrome finish, sharp mirror-like reflection of the surrounding environment, cool neutral highlight with no color tint, studio lighting
[OBJECT] in brushed aluminum, fine parallel brush marks running [DIRECTION], highlights stretched into streaks parallel to the brush direction rather than sharp points
Antique brass [OBJECT], warm golden-toned highlights rather than white, visible patina and darkening concentrated in the recessed detail and edges, soft directional light
Weathered copper [OBJECT] outdoors, green verdigris patina concentrated where rain and weather would actually reach it, bare copper still visible in protected areas, natural daylight
How Do You Prompt Glass Without It Reading as Water or Acrylic?
Glass does two things at once that almost nothing else does: it reflects a little light off its surface and lets most of the rest pass through, bent by refraction, so whatever sits behind it shows up shifted and often magnified. A prompt that only says "glass" and skips both halves usually lands on something either too opaque, reading as white plastic or ceramic, or too invisible, reading as plain water or nothing at all.
Thickness and color deepen together: name where the glass is thickest, because that's where color saturates most, while thin edges read almost clear. "Green glass bottle, color deepening toward the base where the glass is thickest, faint clear rim at the thin edge" gives the model an actual optical rule instead of one flat color. Frosted or etched glass swaps that transmission-plus-reflection pair for pure diffusion: light passes through but scatters into a soft, directionless glow, with no sharp reflection or legible shape visible through it, the opposite failure mode of clear glass and one that needs its own wording rather than "frosted" alone.
Clear glass [OBJECT] on a table, visible refraction bending the [BACKGROUND] behind it, soft reflection of the light source on the surface, color deepening toward the thickest part of the glass
Frosted glass [OBJECT], light passing through and scattering into a soft even glow, shapes behind it reduced to blurred color with no sharp detail visible
Cut crystal [OBJECT] under directional light, small caustic light patterns cast on the surface it's resting on, sharp internal reflections at each facet edge
Thick colored glass [OBJECT] in [COLOR], noticeably deeper and more saturated color where the glass is thickest, near-clear at the thin rim, subtle surface reflection without hiding what's visible through it
Where Do Fabric and Skin Fit Into This Same Framework?
Both sit closer to the diffuse end than metal or glass do. Fabric scatters light across its surface with little to no coherent reflection (silk and satin are the visible exception, with a soft moving highlight rather than a sharp one), and its construction, weave, knit, nap, matters more than any reflective quality. Skin does the same diffuse scattering, plus a small amount of light entering the top translucent layers and re-emerging nearby, which is why thin areas like ears and nostrils pick up a faint glow under strong light instead of staying flat.
Naming construction for fabric, and naming that thin-area glow for skin, answers the same underlying question this whole post is asking, what does this specific surface do with light, just answered differently for each material. Fabric's full material-by-material breakdown, including fur and hair, lives on its own page: Fabric, Fur and Hair: Texture Prompting That Works. This post leaves that ground to it rather than re-teaching it.
Which Image Models Actually Document Material Behavior?
| Feature | FLUX.2 | Nano Banana Pro | GPT Image 2 | Ideogram 4.0 |
|---|---|---|---|---|
| Explicit material-transformation editing (recolor, or change material entirely) | Documented under Color & Material Changes, including hex-code recoloring | Not published as a named feature | Not published as a named feature | Not published as a named feature |
| A vendor worked example uses rendering-pipeline vocabulary (PBR, specular, etc.) | Not found in current docs | Yes, one worked example asks for realistic PBR materials | Not found in current docs | Not found in current docs |
| General guidance on describing materials for photorealism | Documented: product photography guide names materials, reflections and lighting together | Documented: guidance to break complex scenes into steps | Documented: cookbook tells writers to be concrete about materials, shapes and textures | No dedicated materials section found |
Spelled out: Black Forest Labs states plainly that "FLUX generates high-quality product photography with realistic materials, reflections, and lighting" (docs.bfl.ai, accessed September 2026), and separately documents a "Color & Material Changes" editing category, complete with hex-code recoloring and named material-transformation examples. Google's own image-generation guide includes a Nano Banana Pro worked example that asks the model to "Use soft, refined textures with realistic PBR materials and gentle, lifelike lighting and shadows" (ai.google.dev, accessed September 2026), a real vendor-published prompt reaching for rendering-pipeline language, not a keyword the guide defines elsewhere. Neither OpenAI's nor Ideogram's current docs carry an equivalent section naming materials as their own subject.
Is This Physics Vocabulary the Model Actually Understands, or Just Ours?
Most of the words in this post, specular, diffuse, transmission, refraction, Fresnel, subsurface, anisotropic, come from photography and 3D-rendering pipelines, not from any image model's own documentation. Checked directly: none of OpenAI's, Google's, or Black Forest Labs' current prompting guides use "specular," "Fresnel," "diffuse," or "anisotropic" as a defined keyword with a stated effect. The one exception is "PBR," and even that shows up exactly once, inside a single worked example, not as a documented control.
That doesn't make the vocabulary useless. A diffusion model trains on real photographs and rendered images, both of which already encode these physical behaviors, so describing them in plain language, a highlight that keeps the metal's own color, light passing through and bending what's behind it, gives the model something concrete to match even without a special keyword. It's craft convention borrowed from photography and CG, not a documented control any vendor promises will behave a specific way. That's worth saying plainly rather than implying a model "understands" Fresnel reflectance the way it understands an aspect-ratio parameter.
Where Does Wording Stop Working?
Sort the failure the way live post 304's image troubleshooting framework sorts every other image failure, before spending time rewriting a prompt that was never going to fix the problem.
Prompt-fixable: a plastic-looking render because reflection tint and transmission were never named; brushed metal reading as polished because only "brushed" was said instead of naming the streak direction; glass reading as water because nothing described what should refract behind it.
Parameter or setting problem: a reference image of the actual material available but never supplied to a model that accepts one; fine reflection or caustic detail lost to a low output resolution; wanting an exact color match on a model whose docs don't publish that kind of control at all, a genuinely separate question with its own honesty check at the negative prompt support matrix for the equivalent gap on exclusions.
Capability limit: asking a model to hold one exact reflection or caustic pattern identical across regenerations. No model surveyed for this post publishes a reproducibility guarantee for material rendering specifically, Google's own image configuration carries no seed field at all, and where a seed exists elsewhere, vendors describe results as similar, not identical. Treat an exactly repeatable reflection as outside what any of these tools currently promise, not as a prompt you haven't found yet.
Stop rewriting prompts. Start shipping.
Works with ChatGPT, Claude, Gemini, Grok, Midjourney, Ideogram, Veo3 & Kling. 4.8★ on the Chrome Web Store.
Create An AccountCopy-Paste Prompts for Metal and Glass Materials
A few more general-purpose prompts beyond the material-specific ones above. Every bracket is yours to fill in; support for a literal negative-prompt field varies by tool, so the exclusion block below works as "avoid" language stacked into the main prompt rather than assuming a dedicated field exists everywhere, the same distinction the negative prompt support matrix covers in full.
[OBJECT/SUBJECT], [MATERIAL], reflection color and sharpness matching a real [MATERIAL] surface, [LIGHT SOURCE], shot like a real photograph, no plastic sheen
Extreme close-up of a [MATERIAL] surface, [BEHAVIOR: sharp mirror reflection / soft diffused scatter / light passing through and refracting], directional light revealing exactly that behavior
[SUBJECT] holding or wearing [OBJECT] made of [MATERIAL], reflection or transmission consistent with real [MATERIAL], rest of the scene lit to match
Split comparison of the same [OBJECT] in two materials: [MATERIAL A] on the left showing [BEHAVIOR A], [MATERIAL B] on the right showing [BEHAVIOR B], identical lighting on both sides
Append to any existing prompt: add reflection color matching the material's own tone rather than a neutral white highlight, and either a sharp mirror reflection or full transmission depending on the material, removing any generic "shiny" or "glossy" language already present
avoid: uniform plastic sheen, colorless highlight, waxy surface, generic shiny material with no reflection detail