Do I need a professional camera?

No. Amazon’s own seller guide says a newer phone with a 12-megapixel camera or better works for product photos, with a tripod and good light. For AI references, check sharpness and even light before you worry about megapixels.

Can I use product photos I already have?

Yes, if they are sharp, evenly lit and cover the views in the table above. Check them for the usual problems: a label shot at an angle, a close-up taken at 1x, a warm or cool cast from room lights. Then shoot only the views that are missing.

How many reference photos should I send with each image?

As few as the shot needs. Google lists up to 10 high-fidelity object references for Nano Banana 2, and more did not help us. On our surf campaign, the character sheet we kept used 5 references and a short prompt, after attempts with up to 8 references did not work.

How to photograph your product so AI gets it right

Checked September 24, 20266 min read

Short answer

Take a small set of sharp photos that show everything the AI will have to draw: front, back, both sides, top, the label straight on, a close-up of the texture and one photo that shows size. Use soft window light with the lamps off and a plain background, and stand back and zoom to 2x so the phone’s wide lens does not stretch the shape. The model copies what it can see and guesses the rest.

The reference set to shoot

Shoot this set for each product, and again for each color you sell. Name the files by view, such as front.jpg and back.jpg, so the right photo goes to the right shot later.

View	How to shoot it	What the model uses it for
Front	Camera level with the middle of the product, product square to the lens	Shape, proportions and the main face, in nearly every image
Back	Same height and distance as the front	Any image that shows the back. Without it, the back gets invented.
Left and right sides	Same height and distance again	Three-quarter views and depth
Top	From above, lens pointing straight down	Lids, caps and flat-lay images
Label or logo, straight on	Label flat to the lens, filling the frame, no glare	Letter shapes and layout, which models often get wrong
Texture close-up	As close as the phone will focus, light from one side	Weave, grain and finish in close-ups
Small parts	Close-ups of zips, buttons, clasps, pumps and seams	Parts a model tends to add, drop or reshape
Size	In a hand, or next to a common object	The product’s real size in scenes; see how to show product size
As sold	The box, pack or bundle the customer receives	Packaging images and what-is-included images

For clothes, add flat lays of the front and back on a plain floor, and a photo of the garment on a person if you have one. See how to turn a flat lay into on-model photos and how to get back and side views of clothes right.

Why does each view matter to the model?

An image model redraws the product in every image. It keeps what it can see in your photos and fills the rest with a guess based on similar products. A bottle shot only from the front comes back with a made-up back. A jacket with no close-up comes back with a fabric that looks plausible and is not yours. More on this in why AI changes your product.

Models also limit how many references they hold at once. Google’s documentation lists up to 10 object images that Nano Banana 2 keeps at high fidelity, and up to 6 for Nano Banana Pro. OpenAI’s cookbook says the first image in a request keeps the finest detail. Treat the set you shoot as a library, and give each image only the few photos it needs, product first.

When our agency built campaigns by hand, each shot got only the photos it needed: the back photo for a back view, the texture photo for a close-up. One kept prompt for trousers on a model said: "If a fabric close-up reference is provided its weave and texture must be matched exactly."

Light and background

Shoot near a window in daylight, out of direct sun. If sun comes in, hang a thin white sheet or paper over the window, as Amazon’s seller guide suggests.
Turn off lamps and ceiling lights. Room bulbs and daylight are different colors, and mixing them puts a warm tint on one side of the product and a cool one on the other. The model copies both.
Hold a sheet of white foam board on the shadow side to bounce light back in.
For shiny products, move the product or the board until no bright window shape sits on the label. A reflection in the reference can end up in every render.
Use a plain, light background: a white or light gray wall, or a sheet of paper curved from the wall down onto the table so there is no corner line. Patterns and strong colors can tint the edges of the product.
Keep the same light for every view, so the color does not shift from photo to photo.

Why stand back and zoom to 2x?

A phone’s 1x camera is a wide-angle lens. Apple lists it at 24 mm on the iPhone 18 Pro and 26 mm on the iPhone 17, and Nikon classes roughly 14 to 35 mm as wide-angle. Close up, the part of the product nearest the lens looks bigger than it is, because perspective depends on how far the camera is from each part of the object. The model then copies that stretched shape.

Step back and switch to 2x, which every iPhone on Apple’s comparison page offers in September 2026: 48 mm on the iPhone 18 Pro and 52 mm on the iPhone 17. The iPhone 18 Pro also has 4x (100 mm), which suits small products in good light, since that lens gathers less light than the main one. On other phones, use the 2x or longer setting if the camera app has one. Fill most of the frame with the product from the new distance.

Phone settings

Use the rear camera in the standard photo mode. Amazon’s guide notes the rear camera usually has a higher resolution than the front one.
Tap the product to set focus and exposure, then lock them so the phone does not change them between views.
Use a tripod, or rest the phone on something solid at the same height for every view.
Check each photo at 100% zoom. A blurry or tiny reference gives the model less detail to copy, so fill the frame when you shoot instead of cropping a small product out later.
Send the original files, not screenshots. iPhones save photos as HEIF by default, which Gemini accepts. OpenAI’s image input docs list PNG, JPEG, WEBP and GIF, so export a JPEG copy if a tool will not open the file.

Clean the product first

Wipe off dust and fingerprints, and run a lint roller over fabric.
Take off price stickers and shop tags unless they ship with the product.
Steam or press clothes. A crease in the reference can come back in every photo.
Fill soft packs so they are not dented, and turn labels square to the camera.
Use the exact color and version you sell. A sample in a different shade teaches the model the wrong color.

Checklist before you upload

Front, back, both sides and top, all from the same height and distance
Label straight on, readable at 100% zoom, with no glare
Texture and small-part close-ups
One size photo with a hand or a common object
Window light only, with lamps off, on a plain light background
Shot at 2x or longer, never at 1x from close up
Focus and exposure locked, no blur at 100%
Product clean, stickers off, creases out
Files named by view and color
Originals kept, with JPEG copies ready for tools that will not open HEIF

Questions people also ask

Do I need a professional camera?: No. Amazon’s own seller guide says a newer phone with a 12-megapixel camera or better works for product photos, with a tripod and good light. For AI references, check sharpness and even light before you worry about megapixels.
Can I use product photos I already have?: Yes, if they are sharp, evenly lit and cover the views in the table above. Check them for the usual problems: a label shot at an angle, a close-up taken at 1x, a warm or cool cast from room lights. Then shoot only the views that are missing.
How many reference photos should I send with each image?: As few as the shot needs. Google lists up to 10 high-fidelity object references for Nano Banana 2, and more did not help us. On our surf campaign, the character sheet we kept used 5 references and a short prompt, after attempts with up to 8 references did not work.

Where Overs fits

Overs needs one product photo to start. Its shot plan then names the reference pictures each photo needs, in order, such as the character sheet and then the back of the garment, and sends only those to the image model. If a needed picture does not exist, Overs skips it instead of guessing with the wrong one.

Free for 40 photos a month. The AI that makes the photos is billed separately, on your own key, with no markup from Overs.

Start free How it works

Sources

Checked on September 24, 2026. Prices, specs and rules change; follow the links for the current versions.