Product photography prompts: 13 product shot prompts that keep your real product

The formula is a short brief in a fixed order: your product photo, the surface, the background, the light, the camera, a few props, the mood and the use, then one line that keeps the product unchanged. Write it as sentences, not tags. Google's guide to its Gemini image model says it in one line: "Describe the scene, don't just list keywords."

The formula, to fill in:

The keep line, at the end of every prompt:

Two slots decide most of the result. Camera: repeat your photo's angle. The model can only keep what it has seen; ask for a three-quarter view of a product shot head-on and it invents the side. Light: match your photo's light, so the highlights already on the product agree with the scene. Lens words such as "blurred as at f/2.8" describe a look, not physics: OpenAI's guide calls camera settings cues for appearance, not a guarantee.

Without your photo, the model invents a product that only resembles yours. Type "a black coffee bag with a green logo" into an image generator and you get a black coffee bag, not yours: the logo nearly right, the shape slightly off. Fine for a mood board. On a listing, the parcel will not match the picture.

So attach your photo and ask for an edit. ChatGPT, Gemini and product photo tools all accept one. The model then works from your pixels, and the prompt only describes what changes around them.

What the photo has to show

• The whole product, sharp, in soft light. Blur, glare and flash carry into every result.
• The side you want to sell. Ask for a side the photo never showed and the model invents it, back label included.
• The state you want to sell. Ask for a closed jar open and the model makes up the inside.
• Nothing touching it. Price tags and fingers can become part of the product.

You do not need a studio for it. A phone held level, a window to one side and a plain sheet of paper behind the product are enough; the setup is in how to photograph a product with your phone.

Why the keep line works

It is the edit pattern OpenAI's guide recommends: say what changes, list what must stay, and repeat the list on every attempt. Quote the label as printed, and spell an invented brand name out one letter at a time. The line lowers the risk without removing it: small print is still where image models slip most, so check every result at full size.

To see what the formula does with an ordinary photo, we took one: a can of Monster Energy Ultra on a desk, shot on a phone, with a monitor, sunglasses, cables and a pen around it. Each prompt below went through ProduktShot's Photo mode with the default model, one run each at 3 credits an image. Nothing was retouched and no run was repeated.

Photo mode adds the keep line by itself, so these prompts end without it. In ChatGPT or Gemini, add the keep line from the formula to each one.

1. White studio

2. Gym bench

3. Ice on slate

4. Mountain lake

5. Desk flat lay

6. In hand, in the city

7. No prompt at all: the Marble block scene, picked from the scene library.

What held and what to check

• The label held in all seven, down to the small ZERO SUGAR line on the rim. That is the keep line at work, and it is still the first thing to check at full size.
• The white is not pure white. The studio shot looks white, but its corners measure between 244 and 253, not 255. For Amazon's main image, put a cut-out on white instead, as the next section explains.
• One word can change the scene. Our first white prompt asked for a paper backdrop curving up behind the can, and the model drew exactly that: the curve, its edges and the dark studio around it. Asking for no visible edges, corners or horizon gave the clean version.
• The flat lay came out top-down from a front-on photo. A can is the easy case: lying on its side it shows the same label. A box or a bottle with a back label would need its own photo from above.
• The props came in as asked, and nothing else did: the towel and dumbbell, the earbuds and notebook, and not one extra product.

Then we ran the same scene prompts in ChatGPT, with the same phone photo attached and the keep line from the formula at the start of each. ChatGPT has no scene library, so the seventh is the marble block written out as a sentence.

How the two compare

• Both kept every word on the can, ZERO SUGAR on the rim included, and both followed the props and light they were given.
• ChatGPT's white came closer to pure white: its corners measure 254 to 255, against 244 to 253 in our run. Neither is 255 edge to edge, so for Amazon's main image the cut-out advice below still stands.
• Next to the real can, ChatGPT kept the dark outline round the blue ENERGY letters, which our run softened. Small print is where both slip, so compare every result with the product at full size.
• The work around the prompt differs more than the pictures. In ChatGPT each scene is its own message: you paste the keep line into every prompt and download each picture.

Amazon's main image rules are exact: pure white (RGB 255, 255, 255), the product filling at least 85 percent of the image, one unit, no props that are not included, no added text or logos, and no shadow running to the edge. Amazon also says a synthetic image that shows the product realistically, like a good product photo, does not count as an added graphic.

1. Marketplace main image on white

Use it for: Amazon's first image, or any main image that must be plain white.

Change: the margin, if the product comes back small.

Better still, skip the prompt here. A generated background can look white and measure slightly grey, and a soft shadow can creep to the edge. A cut-out placed on white is exact every time: the AI white background generator does that at each marketplace's size and fill. Either way, test the file in the free Amazon image checker.

2. Soft-shadow packshot for your own shop

Use it for: your own shop's product pages, where white is a style, not a rule.

Change: the paper colour. One pale brand tint for every product makes a catalogue look like one shoot.

A lifestyle image answers the buyer's next questions: where does this live, and how big is it? Keep the room believable and every prop smaller than the product.

3. Kitchen counter, morning light

Use it for: food, drink and kitchenware, as a second or third listing image.

Change: the time of day. Warm evening lamp light suits a gift listing.

4. Bathroom shelf

Use it for: skincare, haircare and anything that lives by a sink.

Change: the stone. Travertine feels warmer; black slate feels more expensive.

5. Outdoors, at the real size

Use it for: outdoor gear, drink bottles and garden products.

Change: anything but the size sentence. A wrong size is the mistake a buyer finds only when the parcel arrives.

A flat lay is shot from directly above, so your source photo has to be too: a bottle shot from the front cannot become a top-down view without the model inventing its top. Lay the product down and photograph it from above first.

6. Flat lay with props

Use it for: stationery, accessories and cosmetics sets, on Instagram and Pinterest.

Change: the props. Choose objects your buyer already owns, three at most.

7. Ingredient flat lay

Use it for: food, supplements and natural cosmetics.

Change: the ingredients, but only to ones on your label. Fruit the product does not contain is a claim you cannot back.

Premium and seasonal looks come from light and surface more than from props: one narrow light on dark stone reads as expensive, weather at a window reads as a season.

8. Dark stone

Use it for: perfume, watches, jewellery and premium drinks.

Change: the reflections. Glass and polished metal mirror the room, so ask for black panels on both sides and the edges stay crisp.

9. Seasonal version

Use it for: a holiday campaign without a new shoot.

Change: one thing per run. Swap winter for spring, keep the rest, and the seasonal set still looks like one brand.

A hand shows size better than a ruler. Two Amazon rules first: the main image may not show a person, except for adult clothing, and an image with a photorealistic, entirely AI-generated person needs the keyword contains-synthetic-performer in its IPTC metadata before upload.

10. Hands only

Use it for: small products where size is the question: cosmetics, jewellery, snacks, phone accessories.

Change: the hand, to match your buyer. Then count the fingers.

11. In use, with a person

Use it for: bags, sports gear and accessories, in ads and later listing images.

Change: the viewpoint, to wherever your logo faces. Otherwise the model draws a side it never saw.

Image models now render text, but a few words in clear space are still the safe case. Quote the exact words, say where they go and in what style, and say that nothing else is written. OpenAI's guide adds one step: check the spelling in the result.

12. Square social ad with a headline

Use it for: a feed ad. Keep ad copy off listing images: Amazon allows no promotions or claims in any product image.

Change: the headline, kept short. A few large words are the easy case; a long line of small text is where letters go wrong. Or ask for the empty space and add the words in a design tool.

Start from a finished still, such as the kitchen shot from prompt 3, and describe only what moves. Runway's camera guide says a named move such as "slow dolly push-in" works better than a mood word like "cinematic", and that one move per clip is the reliable fallback.

13. Slow push-in

Use it for: a product page video, a 9:16 story or the first seconds of an ad.

Change: the move: pull back, pan or a partial orbit, but no full 360-degree spin unless you have photos of every side. A negative prompt field takes a list (warped label, extra text), not sentences with "no", as Google's video guide advises.

You do not need a prompt generator. A chat assistant that can see photos fills the formula for you and copies your label text, which saves typos in the keep line. Attach your product photo and paste this:

Check the copied words against the real pack first, because vision models misread small print too. Then run the prompts in the same chat with the photo attached, or in your image tool.

Seven failures cover most bad results. Each row names what you see, why the model did it and what to change in the prompt or the photo.

Before anything goes live, zoom in on every word, find the contact shadow and put the result next to your photo. The full routine is in eight AI product photo mistakes.

ProduktShot's Photo mode runs this method with most of the typing done. A product can hold more than one photo, the hero and other angles, and every Photo prompt carries them. You pick scenes from a library of 103 scenes, filtered by mood and light, and get one image per scene; the prompt field adds a detail or describes your own set.

The keep line is built in. Every prompt Photo mode sends asks the model to keep the product's shape, colour, material, label and logo, and the composer shows this as Product lock. Where ChatGPT takes one scene per message, Photo mode makes one image per picked scene in a single click, at the ratio and size you choose, and keeps every result in your library. It lowers the risk rather than removing it, so open each result and press Compare.

You choose the model (Nano Banana Pro, Nano Banana 2, GPT Image 2.5, Seedream 5 Pro and ProduktShot Lite), the ratio (1:1, 4:5, 3:4, 2:3, 9:16 and 16:9), the size (1K, 2K and 4K) and up to 4 images of one scene per click. An image costs from 2 credits, and the Generate button shows the price before you press it.

Ad posters start from 122 templates and write the headline and offer line for you; read every word before you publish. The presenters you can add are synthetic people, so the Amazon tag applies to those images. You start with 7 credits free, and pictures made before your first payment carry a watermark.