Steampunk Whiskey Bottle: 5 Prompt Fixes for Better
Quick Tip: Copy the prompt with the button above — paste it into Midjourney as-is (--ar and --style raw included). On DALL-E, Flux, Stable Diffusion, Ideogram, Leonardo AI or Firefly, keep the descriptive text and drop the -- flags. Experiment with the background distance first: moving the product away from the backdrop changes the shadow more than any lighting word.
Why Physical Detail Beats Decorative Adjectives in Product Prompts
Here's the strange part: the original steampunk whiskey bottle prompt was already doing most things right. It named lighting angles, Kelvin temperatures, even an f-stop — details most people skip entirely. And yet it stumbled on the one thing that matters most. It described the bottle aesthetically, with phrases like "covered in intricate brass gears," instead of physically, with terms like "raised relief sculptures." That gap between pretty language and physical language is exactly where product prompts succeed or quietly fall apart.
The reason comes down to how diffusion models actually learn. They're trained on photographs, and photographs obey physics. When your prompt names a real physical condition — oxidized copper, a ribbed screw cap, gears with measurable depth — the model can pull from thousands of images where light interacted with that exact surface. Shadows land where they should. Highlights sit where a key light would genuinely place them. Describe something vaguely decorative instead, and the model has to guess, which is how you get gears that look painted on rather than machined.
The improved prompt fixes this systematically rather than cosmetically. Every decorative element gets a construction verb and a location. Gears become "interlocking" and "wrapping the shoulders." The automobile scenes become "bas-relief," a sculptural term with rich training data from architecture and medal photography. Chains are "draped," which implies gravity, weight, and contact points. Each word narrows the geometry the model must render, turning a loose suggestion into a buildable specification — and the difference shows up immediately in the glass.
The Two-Light System: Why Temperature Contrast Sells Metal
Steampunk imagery lives or dies on its metals, and metal has a peculiar optical property: it has almost no color of its own. What we perceive as "brass" or "copper" is largely the color of the light the surface reflects. This is why the original prompt's instinct — a warm 3200K key against a cooler fill — is correct, and why the improved version sharpens it.
Here's the mechanism. A tungsten-balanced key at 3200K hitting polished brass produces rich amber highlights. If the only other light is the same temperature, every reflection on the bottle sits in the same warm family, and the gears, chains, and reliefs merge into a single brownish mass. Introduce a cool source — the improved prompt uses a faint cool rim from behind the right edge — and suddenly the metal surfaces facing away from the key pick up blue-steel reflections. The eye reads this warm/cool split as material complexity: multiple metals, multiple planes, real depth.
The asymmetry matters as much as the temperatures. The key is strong and directional (45 degrees camera-left); the cool light is a rim, not a fill of equal power. Equal-power opposing lights cancel each other's modeling and produce the flat, catalog-style illumination that makes product shots look like clip art. A useful mental ratio: the dominant source should do the storytelling, and the secondary source should only reveal what the first one hides.
One caution on Kelvin values: treat them as relative instructions, not guarantees. The model interprets "3200K" as "warm tungsten character" rather than a metered value. The technique still works because what matters is the gap between sources — the prompt just needs to establish that the key is warm and the counter-light is cool. Prompting a 3200K key against a 3200K fill, by contrast, instructs the model to neutralize the contrast entirely. If you're exploring metal rendering across different generators, Midjourney responds particularly well to explicit Kelvin differentials, while Adobe Firefly tends to interpret lighting language more conservatively — worth testing both with the same prompt to see how each handles the warm/cool split.
Surface, Shadow, and Context: The Product Photography Triad
Professional product photography rests on three decisions: what the product sits on, how its shadow behaves, and what lives behind it. The improved prompt addresses all three deliberately.
Surface interaction. "Polished dark wood surface with soft reflection" does two jobs. The wood grain grounds the bar narrative, and the soft reflection confirms the lighting setup — a reflection is essentially a second, inverted render of the scene, so when it agrees with the bottle's highlights, the whole image gains physical coherence. Omit the reflection and the bottle can appear to float, a common tell of AI product images.
Depth of field as a system. The f/2.8 specification only works because it's paired with "85mm lens perspective" and a background with defined content. Aperture alone is a blur instruction; aperture plus focal length plus subject distance is a look. At 85mm and f/2.8, the bottle stays fully sharp while a bar interior two meters behind dissolves — exactly the rendering the prompt requests with "defocused bar interior with warm bokeh points." Note the bokeh is given a source (lights in the bar) rather than requested abstractly. Bokeh without sources produces generic blur; bokeh with sources produces those characteristic circular highlights at consistent sizes.
Text as texture. The rendered image shows the predictable failure of requested lettering — "JCKDAWELTI'S" instead of the specified brand. This isn't a flaw to fight but a constraint to design around. Diffusion models generate letterforms as visual patterns, not as spelled content. The professional workflow is to prompt the typographic style ("Art Nouveau lettering," "embossed serif") and either accept the pseudo-text at small scale or composite real type afterward. The same principle applies to other ornate product work — the metallic detail strategies in this ultra-luxe golden skeleton prompt rely on the same relief-and-patina logic, and complex surface work like the mosaic fashion portrait succeeds for the same reason: named physical construction instead of decorative adjectives. For darker, texture-heavy material rendering, the grunge street art portrait breakdown shows how oxidation and wear terms anchor surfaces the same way patina does here.
Where to Take This Next
The natural extension of this setup is a controlled variation study: keep the bottle prompt fixed and change one lighting variable at a time — swap the cool rim for a second warm source, drop the key from 45 degrees to near-profile, or move the background from bar interior to seamless dark gray. Each single-variable change teaches you exactly which phrase controls which part of the render, and that mapping is the real asset. Once you know how your generator of choice interprets Kelvin language and relief terminology, the same grammar transfers directly to watches, perfume flacons, and any other product where material and light do the selling.
Key Principle: For AI product shots, replace aesthetic adjectives with physical specifications: name the relief depth, the light's Kelvin temperature and angle, and the lens focal length. Physics is what the model renders convincingly. If you want to go deeper, read golden hour african savanna river.