How to write AI image prompts that look like real product photos

Why AI product images look fake

Three tells give most AI product shots away. First, no contact with the world: the product hovers slightly, casts no shadow, or sits on a surface that does not react to its weight. Second, impossible light: everything is evenly bright, as if lit from everywhere at once, so metal looks like grey paint and glass looks like plastic. Third, over-cleanliness: real product photos have a crease in the cloth, a not-quite-perfect finish, a background slightly out of focus. Generators default to a showroom render unless your prompt asks for the evidence of a real camera in a real room.

The fix is not a longer prompt. It is a prompt that spends its words on what a camera would record: material, light direction, surface, depth of field and one small imperfection.

The five-part formula

Write every product prompt in this order: subject + material + scene + light + camera. One clear sentence is usually enough; a second sentence can add the detail that sells, such as a visible texture or a prop that shows scale. Compare the pair below — same product, same generator, very different brief.

The vague version leaves every decision to the model, and the model chooses the average of everything it has seen: a render. The five-part version decides itself. “Amber glass” tells it how light should pass through the jar. “Worn oak” gives the surface grain to reflect. “From the left” forces one shadow direction, which is what makes an image read as photographed.

If you only remember one row, remember light. One named light source with a direction fixes more “AI look” than any style keyword.

Lighting words that actually change the result

Generators respond well to lighting language borrowed from photography, because their training data is captioned with it. Soft window light gives diffused daylight and soft shadows — the safest choice for food, ceramics, cosmetics and anything handmade. Golden hour warms colours and stretches shadows; good for outdoor products, risky for anything whose colour must be exact. Overcast daylight removes harsh shadows and keeps colours honest, which is why real sellers shoot on cloudy days. Studio softbox gives the clean catalogue look on white or grey. Candlelight creates a cosy mood for evening products, but shifts whites toward orange.

Always give the light a direction — “from the left”, “from behind”, “from a window on the right”. Backlight is the trick behind glowing drink and perfume shots: light shines through the product toward the camera, so edges brighten and liquids glow. And always let shadows exist: “soft shadow under the product” sounds minor, yet it is the fastest cure for the floating-product problem, because a shadow glues an object to its surface.

A worked example, line by line

Here is a complete prompt for a skincare product, with the job each phrase is doing. Steal the structure, swap the product.

Frosted glass sets the material, so highlights spread softly instead of mirroring. Wet slate and water droplets give the surface something to reflect, which sells the spa feeling without a word about spas. The eucalyptus sprig is the scale prop: one familiar object shows the bottle’s size better than any measurement. Light from the right keeps the left side slightly darker, so a headline placed there later stays readable. Nothing asks for text, because text is where generators fail most — more on that below.

If your first result is close but not right, change one part at a time — only the surface, or only the light direction. Changing five things at once teaches you nothing about which word did the work.

Backgrounds, surfaces and ratios

Backgrounds do two jobs: they set the price feeling, and they keep a shop looking like one shop. Pick two or three background recipes and reuse them across every listing. A consistent light, surface and palette across ten products reads as a brand; ten different scenes read as ten different sellers.

Generate at the ratio of the slot the image will fill. A hero composed at 16:9 and cropped to a square thumbnail loses exactly the detail you composed it around. If a marketplace crops thumbnails automatically, keep the product in the middle of the frame: “product centred with space around it”.

Fixing the five failures you will meet first

The plastic look. Everything is too smooth and too shiny. Add a real material with a finish — matte, brushed, unglazed, woven, scratched — and one small imperfection such as a fingerprint-free but slightly uneven rim. Ask for “photograph” language (lens, depth of field) rather than “render” or “3D” words.

Warped or invented text. Generators still misspell, especially long words and small print. Keep in-image text to two or three words, put the exact words in quotation marks in the prompt, and check at 100 percent zoom. For labels, ingredient lists and anything a customer must read, generate the scene without text and add the real text afterwards in your design tool.

Wrong colours. Mood lighting shifts colour. If the exact shade decides the purchase — a lipstick, a yarn, a paint — use neutral wording (“overcast daylight, colours true to life”), avoid golden hour, and compare against your real product before publishing. Where colour is the product, a photograph of the real item still wins.

Extra objects and odd hands. Busy prompts breed clutter: name fewer objects. Hands are the hardest thing generators draw, so if a hand is not essential, frame the shot so none is needed, or use the scale-prop version instead.

A different product every time. When you generate a set, keep the subject and material phrases identical in every prompt and change only the scene words. Save the winning prompt next to the product’s name, so next month’s matching shot takes one edit, not a fresh invention. Our guide to writing a good prompt covers the same structure inside the app, and the library of 60 prompt templates gives you starting points for the most common product scenes.

A repeatable workflow

1. Write the five-part prompt for the hero shot first; it is the image that earns the click.
2. Generate two or three variants and keep the cleanest, not the prettiest — the one with correct edges, honest colour and no stray text.
3. Reuse the same prompt skeleton for the lifestyle, close-up and scale shots, changing only scene and camera words.
4. Clean up, then finish: remove the background if a marketplace slot needs white, using the remove-background guide, and upscale only at the end if the slot needs more pixels.
5. Check at full size before publishing: text, hands, reflections and the product’s edges, in that order.

In Movapix an image costs about $0.008, charged only when a generation succeeds, so a four-image set with retries still costs a few cents — the full maths is in our guide to product photos for a small shop. Creations are private by default, credits never expire, and downloads may be used commercially. Selling on Etsy? Read Etsy listing photos with AI next: the rules there decide what you may publish.