Skip to main content
Prompt Guides

How to Write Prompts for AI Product Photos

A simple structure for AI product photo prompts, with examples for hero shots, lifestyle scenes, labels and photo edits, plus the mistakes that waste credits.

A good product photo answers three questions at a glance: what is it, what does it look like up close, and why would I want it? An AI product photo prompt has to answer the same questions, only in words. The clearer your words, the fewer generations you need before you get something usable.

This guide gives you a structure you can reuse for almost any product, with examples you can copy and adapt.

Start with the product, not the mood

The most common mistake is to start with atmosphere: "a beautiful, premium, luxurious image of a skincare bottle". Words like beautiful and premium do not tell the model what the bottle looks like.

Describe the product first, as if you were briefing a photographer who has never seen it:

  • Shape and material: a frosted sage-green glass dropper bottle, a straight-sided plastic bottle, a matte aluminium can.
  • Colour: name the colours of the body, the cap and the label.
  • Label: what it says, in quotes, and the style of the lettering.
  • Size cues: in a hand, next to a cup, on a shelf.

For example: a frosted sage-green glass dropper bottle with a matte white cap and a small minimal label reading "VERDA" in a thin sans-serif. That one sentence already fixes most of what matters.

A five-part structure

Once the product is described, add the rest in this order:

  1. Product — what it is and what it looks like (above).
  2. Setting — where it sits: a white seamless studio, stacked travertine blocks, a bathroom shelf, a supermarket aisle.
  3. Light — soft daylight from the left, hard shadows, a warm rim light, cinematic studio lighting.
  4. Camera — close-up, three-quarter view, top-down, macro on a detail, or a lens look such as "100mm macro".
  5. Format and limits — the aspect ratio (1:1 for marketplaces, 4:5 or 3:4 for social posts) and what not to include: no extra text, no logos, no watermarks.

Put together:

Hyper-realistic commercial product photography, 3:4. A deep matte-blue plastic protein drink bottle of a fictional brand, with an original label reading "PEAKLINE", covered in condensation droplets, three-quarter view. A creamy milk splash swirls around the base, with blueberries and mint leaves frozen in mid-air. Background gradient from dark blue to warm amber with soft bokeh. Cinematic studio lighting, all label text spelled exactly as given.

You do not need every part every time, but going through the list stops you from forgetting the one detail that would have fixed the image.

Four types of product shot

Most shops need the same handful of images. Here is how the prompt changes for each.

1. The clean hero shot

The main image on a product page. Keep the background simple and the product sharp.

Front-facing hero shot of [product] in a clean studio, neutral light-grey seamless background, soft even lighting, gentle contact shadow, 1:1, no props, no extra text.

2. The detail close-up

Shows texture, materials and quality.

Extreme close-up of [the label / the stitching / the glass texture] of [product], shallow depth of field, soft side light that brings out the texture.

3. The lifestyle scene

Shows the product in use or in its natural place. Name a few props, not ten.

[Product] on a refined bathroom shelf with folded linen and a ceramic dish, morning daylight from a window on the left, calm and minimal.

4. The exploded or "how it works" view

Good for gadgets and anything with interesting parts.

Exploded view of a wireless over-ear headphone, generic design with no logos, with the ear-cup shells, drivers, cushions, circuit board and battery suspended in alignment above each other. White seamless background, ultra-realistic macro product photography.

Getting labels and text right

Text on packaging is easy to get wrong, and a misspelled label ruins an otherwise good image. A few habits help:

  • Quote the exact words. A label reading "PROTEIN SHAKE - BLUEBERRY YOGURT".
  • Keep it short. A brand name and one product line are easier to get right than a full ingredients list.
  • Say "no other text". Models sometimes add slogans or small decorative words. Add a line such as do not add any extra slogans or decorative text; every word must be spelled correctly.
  • Check every letter before you use an image. Zoom in on small text.

Use your own brand, not someone else's

Only put your own brand, or a made-up one, on the product. Do not ask for another company's logo, packaging or bottle shape, and do not describe a famous product so closely that it looks like theirs. Apart from the legal risk, a mockup that looks like a competitor does not help you sell your own product. The examples in this guide all use fictional brands for the same reason.

Editing a photo you already have

If you already have a decent photo of your product, image-to-image can be quicker than starting from scratch. Upload the photo and describe only the change:

Keep the bottle and its label exactly as they are. Replace the background with a pale stone surface and eucalyptus sprigs, soft daylight from the left.

Always say what must stay the same. Without that line, the model may "improve" the product too, and a product photo that does not match the real product is a problem for your customers.

Mistakes that waste credits

  • Stacking adjectives instead of details. "Stunning, amazing, ultra-premium" adds little. One concrete detail is worth more.
  • Too many props. Three well-chosen props beat a crowded table.
  • Forgetting the format. If you need a square marketplace image, say 1:1.
  • Changing everything between attempts. Change one thing per try so you know what helped.

Try it

On image2ai, both GPT Image 2 and Seedream 5.0 Flash support text-to-image and image-to-image. During the trial, each image costs 5 credits, and verifying your email gives you 10 credits to start. Trial images are watermarked and cannot be downloaded, so use them to test your prompt before you commit.

Start from a written prompt on text-to-image, or upload a product photo on image-to-image. Results vary from prompt to prompt, so it is worth running the same prompt on both models and keeping the one that fits your product best.