Product Shot Prompts: Angles, Surfaces, Light

A product photographer charges for three decisions: where the camera sits, what the product sits on, and where the light comes from. Everything else is execution. AI product shots fail for the same reason cheap product photos fail — those three decisions were never made, so the model makes them for you, and the model's default taste is "generic packshot from a 2019 stock library."

This guide is a vocabulary for making those three decisions explicitly in your prompt: angles, surfaces, and light, plus the composition patterns that separate a scroll-stopping product shot from filler. It builds on the general style keyword testing approach but goes deep on the one genre where specificity pays fastest.

Angles: name the camera position or lose control of it

"Product photo of a serum bottle" leaves the camera anywhere. Professional product photography uses maybe six angles, and naming them works because models learned them from millions of ecommerce listings:

• Straight-on, eye level — the Amazon listing default. Honest, symmetrical, boring in the good way.
• Three-quarter hero (15–30° above) — the workhorse. Shows the face and top of the product, adds dimensionality.
• Low-angle hero — camera below product, shooting up. Makes bottles and boxes monumental; the "premium" cheat code.
• Top-down flat lay — product plus props arranged on a surface. The lifestyle and social-first angle.
• Macro detail crop — texture, cap threads, label embossing. Use as the second image in a set, never the first.
• Dutch tilt — slight rotation for energy. Sparingly, for sale banners and social ads.

Prompt pattern: lead with the product, then the angle as a photographic phrase — "matte black serum bottle, low-angle hero shot, camera slightly below product level." On Flux 1.1 Pro these angle phrases are honored consistently enough that you can plan a whole listing set around them.

Surfaces: the most underwritten line in product prompts

The surface is half the shot's personality, and most prompts never mention it. Each surface implies a positioning story:

Surface keyword	Look it produces	Brand signal
white seamless sweep	Shadowless, floats in space	Marketplace listing, clinical trust
brushed concrete slab	Cool gray texture, soft shadow	Modern, unisex, "design object"
travertine / marble pedestal	Stone veining, luxury plinth	Premium skincare, fragrance
warm oak tabletop	Domestic, tactile	Food, craft, family brands
colored acrylic with reflection	Glossy mirror-line under product	Y2K, playful, social-first
crumpled linen	Soft folds, lived-in	Wellness, organic, slow living
wet slate with droplets	Dark, glistening	Grooming, "active ingredient" energy

Two execution notes. First, say how the product meets the surface: "sitting on," "half-buried in," "casting a soft shadow onto." Contact shadows are what sell physical presence — if the product looks pasted on, the render reads fake regardless of resolution. Second, match surface scale to product scale; a lipstick on a vast marble slab reads lost unless you crop tight.

Light: three setups cover 90% of product work

Skip mood adjectives. Product lighting is a small set of named setups, and models know them:

1. Soft wraparound (big softbox both sides) — even, shadow-light, ecommerce-safe. Prompt: "soft diffused studio lighting, gentle shadow under product."
2. Single hard key with deep shadow — one strong light from the side, dramatic falloff. Prompt: "single hard light from upper left, long crisp shadow, dark background." This is the "editorial" look that makes a $12 product look $60.
3. Backlit glow — light behind translucent products (beverages, serums, candles). Prompt: "backlit, light glowing through the liquid, rim highlights on the bottle edges."

Add one specular detail for realism: "specular highlight running down the glass" or "soft gradient reflection on the metal cap." Reflective and transparent materials are where AI product shots most often crumble, and explicitly describing the reflections is the fix that costs nothing.

Composition patterns for a full listing set

One image is never the deliverable — a listing or campaign needs a set. A structure that works:

1. Hero — three-quarter angle, brand-signal surface, setup #1 or #2.
2. Detail — macro crop of the label or texture, same lighting family.
3. In-context — flat lay or in-hand scene with 2–3 props that reinforce the ingredient story ("surrounded by fresh citrus halves and green leaves").
4. Graphic/social — colored backdrop, acrylic reflection, bolder light.

Keep surface and lighting vocabulary consistent across the set and it reads as one shoot, not four lucky generations. When a hero frame nails the look but the label came out mangled — the most common failure — fix it with inpainting in the AI photo editor rather than re-rolling the whole composition and hoping.

From still to motion without reshooting

The same discipline pays twice: a clean hero shot is the ideal source frame for a product video. Feed your best still into the AI product video generator and prompt only the motion — "slow 180° orbit," "camera pushes in as backlight blooms," "droplets slide down the glass." Because the angle, surface, and light were already decided in the still, the video inherits an art-directed look instead of gambling on a text-to-video roll. One good prompt, one good frame, an entire listing plus a Reel.

FAQ

Why do my AI product shots look like generic stock photos?

Because the prompt didn't make the three decisions that define product photography: camera angle, surface, and lighting setup. Left unstated, models default to the most common look in their training data — straight-on, white background, flat light. Name all three explicitly and the generic look disappears immediately.

How do I keep my actual product accurate in AI shots?

Pure text-to-video and text-to-image will invent a plausible product, not yours. For real-SKU work, start from your own photo: use image editing and inpainting to change the surface and lighting around the product, or generate the scene and composite your packshot in. Reserve fully generated products for concept and mood work where label fidelity doesn't matter.

What resolution and aspect ratio should I generate for ecommerce?

Generate at the model's native quality and upscale to 4K afterward rather than forcing giant dimensions in one pass. For aspect ratios: square (1:1) for marketplace grids, 4:5 for Instagram feed, 9:16 for Stories and product Reels. Generating the hero in 4:5 and cropping to 1:1 usually beats generating both separately.

Can AI handle reflective products like glass and metal?

Yes, but only if you describe the reflections instead of hoping. Phrases like "specular highlight along the bottle shoulder," "soft window reflection in the chrome cap," and "backlit glow through amber liquid" give the model a physical spec to satisfy. Unprompted, reflective materials are where renders most often look plastic.

Should I use the same prompt across different image models?

Keep the structure, expect to retune the vocabulary. Angle and lighting phrases transfer well between models; surface textures and material realism vary more. Run your hero prompt on two or three models before a big batch — the differences in glass, metal, and liquid rendering are big enough to change which model you ship with.

Build your next listing set in Versely: generate heroes in the text-to-image tool, repair labels with inpainting, and spin the winning frame into a product video — all on one credit balance.