Do I need to know camera settings or lighting theory to write better product prompts?

No, but you need to know the vocabulary photographers use. You don't need to own a light meter, but you do need to know the difference between "key light," "fill light," and "rim light," and where each sits relative to the product. That's learnable in 10 minutes and is the missing piece in most AI prompt guides.

Why do my product photos look realistic sometimes and fake other times with the same prompt?

AI generation has variance built in. Your prompt is a direction, not an instruction. If your prompt is vague, that variance is huge—you get wildly different lighting, backgrounds, and product angles. The more specific you are with lighting keywords and background formulas, the narrower that variance becomes. Consistency requires constraint.

Can I use the same lighting setup for jewelry, skincare, and apparel photos?

No. Jewelry needs rim lighting to show shine and facets. Skincare needs shadow softness to show bottle shape and label clarity. Apparel needs side fill light to show texture and fit. The primary source angle stays similar, but modifiers, shadow depth, and fill ratios change by product type. That's why category-specific templates work—they're built on these differences.

How many times do I usually regenerate before I get a usable shot?

With vague prompts, 8–15 times. With precise lighting and background formulas, 2–3 times. The difference is that the first two generations fail for identifiable reasons (shadow too deep, background wrong depth) that you fix in one line, instead of starting over.

Does this work for both Midjourney and DALL-E?

Yes. Both respond to specific lighting keywords and structural prompt layers. DALL-E is slightly more literal with material descriptions, Midjourney slightly more responsive to reference aesthetics, but the core 9-layer anatomy works on both. Parameter syntax differs, which is why a quick reference guide matters.

What to Type for AI

Why Your AI Product Photos Look Fake (And How to Fix It)

Name: AI Product Photo Prompts: 12 Scene Templates + Lighting Formulas
Brand: What to Type for AI
SKU: WTTF-9RHCADNSNAIHSHEMKHUBUG
Price: 29 USD
Availability: InStock

You've generated 50 product images in Midjourney and they look nothing alike. One has harsh shadows that hide your product. Another is drowning in soft light with no dimension. A third has the background so blown out your item disappears into it. The problem isn't the AI—it's that product photography lighting has rules, and those rules aren't in generic prompt guides.

Lighting in product photos isn't abstract. It's a specific chain of decisions: where the light comes from, what it bounces off, how shadows fall, what temperature it carries. AI generators respond to this precision the same way a real camera does. Miss those keywords and you'll waste credits chasing inconsistency. Get them right and every variant of your product—different color, different angle—will look like it belongs in the same catalog.

This is the core problem ecommerce sellers hit when they move from hiring photographers to generating images themselves. You need to think like a lighting gaffer, not a writer.

AI Product Photo Prompts: 12 Scene Templates + Lighting Formulas

$29

Pay once. Keep forever.

Stop wasting API credits on product photos that look like stock images. This prompt guide gives you 10 battle-tested scene templates — each built for a specific product category — so your first or second AI generation is usable, not a starting point. Every...

One self-contained PDF. No hidden files or separate templates.

What's included

10 complete, copy-paste prompt templates covering beauty flat-lay, apparel on model, coffee/beverage, tech gadget, home décor shelf, jewelry with hand, skincare minimalist, candle ambient, eyewear portrait, and accessories flat-lay
Lighting keyword chains for each category — specific primary source, modifier, and mood terms that work across 500+ brand styles
Background formulas: material + color descriptor + depth cue for every template
Hand-placement and pose triggers that prevent the stiff mannequin problem in model and jewelry shots
Before/After failure analysis for every template — exactly which layer to fix and why
Universal 9-layer prompt anatomy builder for any product category not covered by the templates

Get it — $29 →

Or get free updates & new releases:

Follow for updates

The Three Layers That Break Most AI Product Photos

Bad product photos usually fail in one of three places, and each one is fixable once you know what to look for. **Lighting source is vague.** Prompts that say "professional lighting" or "studio light" are too generic. AI defaults to centered, flat illumination. Real product work uses directional setups: "key light from upper left at 45 degrees," "fill light bouncing from white card," "rim light separating product from background." You need to name the specific angle and source, not hope the model guesses your intent. **Shadows disappear or dominate.** Either your product has no definition—it looks rendered and plastic—or shadows are so dark they hide detail and color. The fix is a one-line addition to your prompt that controls shadow depth and falloff. This works across every product category and takes one edit to test. **Background reads as competing.** Your watch sits on a surface you can't identify. Your skincare bottle is lost in murky beige. Backgrounds need a formula: material type + color + how far back it recedes. Without that structure, the background becomes visual noise instead of context.

How Lighting Keywords Stack in Product Prompts

Product lighting works as a chain. Each element does one job. Start with the primary source: "soft north light," "tungsten key light," "harsh midday sun" (this controls mood and direction). Then add the modifier: "bounced through white fabric," "gelled warm," "with deep shadows" (this controls how the light behaves). Then anchor it to a real reference: "like a high-end jewelry advertisement," "cosmetics counter display," "lifestyle brand lookbook" (this tells the model what professional standard to match). Without this layering, you get muddy results. With it, you can regenerate the same product in five different lighting scenarios and they all read as intentional, not random. The specific keyword chains for beauty flat-lay differ from apparel-on-model differ from jewelry-with-hand—each has different shadow tolerance, background depth, and hand-placement needs. Once you map those chains, you stop burning credits on test shots and start shipping images that convert.

The Diagnostic Step Most Sellers Skip

When a generation fails, most people just delete it and regenerate. That's credit waste. Instead: pull the image aside and ask one of three questions. Is the problem in the lighting (shadows wrong, product too flat, weird hot spots)? Is it in the background (too busy, wrong depth, distracts from product)? Or is it in the object itself (pose is stiff, hand placement looks mannequin-like, product angle is unflattering)? Each problem has a one-line fix. Lighting issue? You adjust the source direction and shadow depth keywords—nothing else changes. Background problem? You swap the material and depth cue—keep the lighting intact. Object problem? You insert a specific pose or hand-placement trigger. This is faster than regenerating from scratch, and it trains you to see what actually works in your category. After five or six fixes, you stop guessing. You know exactly which keywords control what, and your generation success rate jumps from 30% to 80%.

Why Consistency Breaks Across SKUs and Variants

You nail the first product photo. You run the same prompt on your second color variant and it looks like it was shot in a different studio three years later. This happens because you're missing the 9-layer anatomy of a working product prompt. Most sellers write: "skincare bottle, minimalist, clean, professional." That's 4 layers. A prompt that regenerates consistently has 9: object + material + pose/angle + primary lighting + shadow control + background material + background color + background depth + reference aesthetic. Change any layer and the image shifts. Lock all nine and variants stay coherent. The other consistency killer is vague references. "Professional" means nothing to AI. "Like Glossier's product page photography, flat-lay with product angled 20 degrees left, white paper backdrop" is specific enough to repeat. Specific references—brand names, exact angles, material names—are the difference between "I hope this works" and "this will work."

FAQ

Do I need to know camera settings or lighting theory to write better product prompts?: No, but you need to know the vocabulary photographers use. You don't need to own a light meter, but you do need to know the difference between "key light," "fill light," and "rim light," and where each sits relative to the product. That's learnable in 10 minutes and is the missing piece in most AI prompt guides.
Why do my product photos look realistic sometimes and fake other times with the same prompt?: AI generation has variance built in. Your prompt is a direction, not an instruction. If your prompt is vague, that variance is huge—you get wildly different lighting, backgrounds, and product angles. The more specific you are with lighting keywords and background formulas, the narrower that variance becomes. Consistency requires constraint.
Can I use the same lighting setup for jewelry, skincare, and apparel photos?: No. Jewelry needs rim lighting to show shine and facets. Skincare needs shadow softness to show bottle shape and label clarity. Apparel needs side fill light to show texture and fit. The primary source angle stays similar, but modifiers, shadow depth, and fill ratios change by product type. That's why category-specific templates work—they're built on these differences.
How many times do I usually regenerate before I get a usable shot?: With vague prompts, 8–15 times. With precise lighting and background formulas, 2–3 times. The difference is that the first two generations fail for identifiable reasons (shadow too deep, background wrong depth) that you fix in one line, instead of starting over.
Does this work for both Midjourney and DALL-E?: Yes. Both respond to specific lighting keywords and structural prompt layers. DALL-E is slightly more literal with material descriptions, Midjourney slightly more responsive to reference aesthetics, but the core 9-layer anatomy works on both. Parameter syntax differs, which is why a quick reference guide matters.