Product photos without a studio.

AI product photography works best when it starts from a real photo of the real product. Shoot the item once in good light, cut it out, place it on pure white for your main marketplace image, and let AI build lifestyle scenes around it for the rest. The product’s shape, color, label and size stay exactly as they are.

Generating the product itself from a text prompt is the mistake to avoid. Image models redraw labels, change proportions and invent details, and marketplaces require images that accurately show what you sell.

01Start from a real photo, not a prompt

Your photo is the evidence of what the customer will receive, so it has to be good enough that the product itself never needs fixing. A phone, a window and a sheet of white card are enough:

• Soft, even light. Shoot beside a window on an overcast day, and bounce light into the shadows with white card.

• A steady camera. Use a tripod, and turn off beauty or “enhance” modes.

• A plain background that contrasts with the product, so cutouts come out clean.

• Every angle a shopper would ask about, plus close-ups of details and one shot in a hand for scale.

• True color. Compare the photo with the real item, and correct the color before any AI step.

02What are Amazon’s product image requirements?

Every marketplace sets its own rules, and the main image, the one shoppers see in search results, gets the strictest ones. This is what the official pages say as of September 2026.

Amazon’s full Product image guide in Seller Central has more detail.

Google is the most explicit about AI. Merchant Center says every image created with generative AI must carry metadata that says so, such as the IPTC DigitalSourceType value TrainedAlgorithmicMedia, and you must not remove embedded metadata. Some export settings strip it, so check your final files.

03Make a compliant main image in one command

Start from a cutout: a transparent PNG from any background remover. This ImageMagick command trims empty space, scales the product to 1,700 pixels on its longest side, and centers it on a 2,000 by 2,000 pure white square. Since 1,700 is 85% of 2,000, the product fills exactly the share Amazon asks for.

magick cutout.png -trim +repage -resize 1700x1700 \
  -background white -gravity center -extent 2000x2000 \
  -alpha remove -alpha off -quality 92 main.jpg

Then check the result before you upload. This short Python script uses Pillow to test three rules you can measure: size, a pure white edge and fill.

import sys
from PIL import Image

path = sys.argv[1]
img = Image.open(path).convert("RGB")
w, h = img.size
px = img.load()

# 1. Size: Amazon prefers more than 1,000 px on each side, so shoppers can zoom
print(f"size: {w} x {h} px", "OK" if min(w, h) > 1000 else "SMALL (no zoom)")

# 2. Background: every pixel along the outer edge should be pure white (255, 255, 255)
edge = [px[x, y] for x in range(w) for y in (0, h - 1)] + \
       [px[x, y] for y in range(h) for x in (0, w - 1)]
off = sum(1 for p in edge if p != (255, 255, 255))
print(f"edge pixels not pure white: {off}", "OK" if off == 0 else "FIX")

# 3. Fill: how far the product stretches across the frame, on its longest side
box = Image.eval(img.convert("L"), lambda v: 255 if v < 240 else 0).getbbox()
if box:
    fill = max((box[2] - box[0]) / w, (box[3] - box[1]) / h)
    print(f"product spans {fill:.0%} of the frame", "OK" if fill >= 0.85 else "TOO SMALL")

Real output for that image, and for a test image with a small product on an off-white background:

$ python3 check_main_image.py main.jpg
size: 2000 x 2000 px OK
edge pixels not pure white: 0 OK
product spans 85% of the frame OK

$ python3 check_main_image.py bad-main.jpg
size: 900 x 900 px SMALL (no zoom)
edge pixels not pure white: 3600 FIX
product spans 31% of the frame TOO SMALL

We tested both with ImageMagick 7.1.2, and with Pillow 9.5 and 12.2. The fill check is a rough guide: a white product on a white background can fool it, so trust your eyes too.

04Add AI backgrounds without changing the product

Lifestyle images, which show the product in use or in a setting, are where AI earns its keep. There are two ways to make them.

1. 01

   Composite: the safest method

   Ask the model for an empty scene with space left for your product, then place your cutout on top and add a soft shadow. Every product pixel is still your photo.

2. 02

   Edit around the product: faster, but check it

   Give the model your photo as a reference and ask for a new setting. The model redraws the product, so compare the label, the proportions and the color with the original. A masked edit, where you mark the area to change, keeps the product safer, though some models treat the mask only as a guide.

Either way, make the scene’s light come from the same side as the light in your photo. Mismatched shadows give a fake away fastest.

A square lifestyle product photo background with no product in it: a pale travertine bathroom shelf beside a window, a folded linen towel at the far left edge.
Soft morning light from the left, with gentle shadows falling to the right. Leave the center of the shelf empty for a perfume bottle about 12 cm tall, seen at eye level.
Photographic, neutral colors, no text.

Place the product from the attached photo on a pale travertine bathroom shelf in soft morning light from the left.
Do not change the product in any way: keep its shape, proportions, colors, label text and cap exactly as in the photo.
Add a soft, realistic contact shadow under it. No other products, no text.

To choose a model for the job, see the best AI image generators of 2026. For the camera and light words that make a scene look photographed, see how to write AI image prompts that look professional.

05Keep lighting consistent across your catalog

A catalog looks professional when every product seems to come from the same shoot. Write the rules down once and paste them into every scene prompt.

House style for all product scenes: key light from the left at about 45 degrees, soft and diffused; gentle shadows falling right; warm neutral palette of cream, sand and pale wood; eye-level camera, square crop, product centered with generous space around it; no text, no people, no competing products.

Pair it with a fixed shot list, such as front, three-quarter view, back, a detail and one scene, so shoppers comparing two products see the same angles.

06The tools that work

What each tool does, as of September 2026:

07Honest product photos: what you can and cannot change

The test is simple: the photo must match what arrives in the box.

This is not just good manners. In the US, the FTC says advertising must be truthful and not deceptive, and it judges an ad by its “words, phrases, and pictures” together. An edited photo that oversells a product can mislead as much as a false claim. Google and eBay both require photos that accurately show the item.

For more ways a small shop can use AI safely, see 15 jobs a small business can hand to AI.

08FAQ

Read next: AI for marketing, a practical playbook for small teams.