AI Product Photos for Amazon, Flipkart and Meesho Sellers (Without Faking the Product)

By Ravindra Bagale, trainer · LinkedIn · Last updated: 23 October 2026

Generative AI course · Chapter 25 · Estimated reading time: 12 minutes

1. Suppose we sell steel bottles from Rajkot

Suppose we run Shakti Steelware in Rajkot. Our best product is a 1-litre stainless steel water bottle with a blue cap and our printed logo. We want to list it on Amazon, Flipkart and Meesho.

We take a photo on the kitchen counter and ask an AI tool:

Make this a professional product photo for Amazon

The result looks great: a shiny bottle in a studio. But look closely. The cap is now dark grey, the logo text is blurred, and the bottle looks taller and thinner. A customer who receives our real bottle will feel cheated.

2. The golden rule

AI may change the background, the light and the scene. It must never change the product: its colour, shape, size, logo, text, parts or what is in the box.

Write this rule into every prompt:

Keep the product exactly the same: same colour, shape, proportions, cap, logo
and printed text. Do not redraw or "improve" the product.

And after every result, compare the product with the real one in your hand.

3. Amazon's main image rules, in simple words

Amazon's seller help pages say the main image (the first image, shown in search results) must:

• Have a pure white background (RGB 255, 255, 255).
• Show the product filling at least 85% of the image.
• Be 1,000 pixels or more on the longest side to allow zoom (minimum 500, maximum 10,000).
• Show the actual product, not a drawing, and the whole product, not cut off.
• Have no text, logos, watermarks, borders or extra graphics added on top (the product's own printed logo is fine).
• Show no props or accessories that are not included.

Some categories (clothing, shoes, jewellery) have extra rules. Flipkart and Meesho publish their own image guidelines in their seller panels. Check those pages for your category before you upload.

4. Take a phone photo that AI can work with

AI cannot fix a bad start. A good starting photo takes 5 minutes:

1. Place the product on a plain white sheet or chart paper, near a window in daylight.
2. Switch off the tube light (mixed light gives a yellow or blue tint).
3. Hold the phone straight, at the product's middle height.
4. Fill most of the frame with the product, and keep it sharp (tap to focus).
5. Take 3 photos: front, side and back (or label).

5. Worked example 1: a white-background main image

Prompt (in Gemini Create images or ChatGPT, with your photo uploaded):

This is a photo of my product, a 1-litre stainless steel bottle with a blue cap.
Keep the product exactly the same: same colour, shape, proportions, cap, logo
and printed text. Do not redraw or "improve" the product.
Change only the background to pure white (#FFFFFF), with no shadow on the
background. Centre the bottle so it fills about 90% of the image height.
Square 1:1, high resolution.

Expected output (yours will be similar, not the same): the same bottle on a pure white background, centred and large.

Check 1, by eye: hold the real bottle next to the screen. Same blue? Same logo? Same cap shape?

Check 2, watermarks: images made in the Gemini app carry a visible watermark. A main image must not have a watermark, so check the corners of every AI image before you use it. If your tool adds a visible mark, use a different tool or method for the main image. Many sellers simply shoot the main image on a white sheet and use a background remover, which keeps the real product pixels and avoids the problem.

Check 3, with Python: see section 8.

6. Worked example 2: a lifestyle image for the second slot

Images after the main one can show the product in use. This is where AI scenes help most.

Prompt:

Keep the product exactly the same: same colour, shape, proportions, cap, logo
and printed text. Do not redraw or "improve" the product.
Place the bottle on a wooden office desk next to a laptop and a notebook,
morning light from a window, shallow depth of field.
The bottle stays the main subject, at its real size compared with the laptop
(it is 27 cm tall). Square 1:1. No text.

Expected output (yours will be similar, not the same): an office desk scene with our bottle sharp in front and the laptop softly blurred behind.

Check size: the real bottle is 27 cm tall. Next to a laptop, it should look roughly as tall as the laptop screen is high. If the AI made it look like a small flask, say: "Keep everything, but show the bottle at its real size: 27 cm tall, about as tall as the laptop screen."

7. Worked example 3: plan a set of 5 listing images

Ask AI to plan the set and write the text for an infographic, then check every claim.

Prompt:

Plan 5 listing images for an online marketplace for this product:
1-litre stainless steel water bottle, blue cap, keeps water cold for
[HOURS - I will measure], leak-proof cap (tested by us), weight 290 g.
Image 1 must be a pure white main image. For each image give: purpose,
what to show, and any text (max 6 words per line). Use only the facts given.
Mark any fact I have not given as [CHECK].

Expected output (yours will be similar, not the same):

Two important points:

1. Image 3 is a real photo. Size in the hand is a promise about scale; take it with your phone, do not generate it.
2. Replace [HOURS] only after you test it. Fill the bottle with cold water in the morning and check in the evening. Never write "24 hours cold" because a competitor does.

Add the text to images 3, 4 and 5 in a photo editor, so the spelling and numbers are exactly right.

8. Check a main image with Python

This short script checks 3 things: the size, a pure white background at the corners, and how much of the image the product fills.

# Check an Amazon-style MAIN image: size, pure white background, product fill.
import sys
from PIL import Image, ImageChops

def check(path):
    im = Image.open(path).convert("RGB")
    w, h = im.size
    px = im.load()
    corners = [px[0, 0], px[w-1, 0], px[0, h-1], px[w-1, h-1]]
    white = all(c == (255, 255, 255) for c in corners)
    # bounding box of everything that differs from the corner colour
    bg = Image.new("RGB", im.size, corners[0])
    diff = ImageChops.difference(im, bg).convert("L").point(lambda v: 255 if v > 10 else 0)
    box = diff.getbbox()
    fill = max((box[2]-box[0]) / w, (box[3]-box[1]) / h) if box else 0
    print(path)
    print(f"  size {w} x {h}  -> {'OK' if max(w, h) >= 1000 else 'too small for zoom (need 1000+)'}")
    print(f"  corners pure white -> {'OK' if white else 'NO ' + str(corners[0])}")
    print(f"  product fills {fill*100:.0f}% -> {'OK' if fill >= 0.85 else 'too small (need 85%+)'}")

for p in sys.argv[1:]:
    check(p)

Run it:

1. Install Pillow once: pip install pillow
2. Save the script as check_main_image.py.
3. Run: python check_main_image.py bottle-main.png bottle-phone-photo.png

Real output (we ran it on 23 October 2026 on two test images: a cleaned 2000 x 2000 main image and an 800 x 800 phone photo on a cream counter):

bottle-main.png
  size 2000 x 2000  -> OK
  corners pure white -> OK
  product fills 90% -> OK
bottle-phone-photo.png
  size 800 x 800  -> too small for zoom (need 1000+)
  corners pure white -> NO (248, 246, 240)
  product fills 50% -> too small (need 85%+)

The phone photo fails all the checks that matter: it is too small, the "white" counter is really cream (248, 246, 240), and the bottle fills only half the frame. Note that a JPG file can turn pure white into 254 or 253 because of compression, so save the main image as PNG, or as a high-quality JPG and check again.

9. Claims, watermarks and honesty

• No claims you have not tested: "BPA free", "24 hours cold" or "food grade" only if you have proof from your supplier or your own test.
• No other brands' logos or "compare with" pictures of a competitor's product.
• No AI watermark on the main image, and no added text on it.
• Colour must match: if your cap is navy, the photo must show navy. Colour mismatch is one of the most common reasons for "not as described" returns.
• Variants need their own photos: if you sell red and blue caps, photograph each one; do not ask AI to "recolour" a cap you do not stock.

10. Worked example 4: white background with remove.bg

Suppose Shakti Steelware in Rajkot needs Amazon main images for five steel bottles. Gemini redraws can change the blue cap shade. For the main image, keep the real product pixels and only remove the background. We use remove.bg (high India traffic for this job; plans and free credits change — check remove.bg before a big batch).

Steps:

1. Open remove.bg in your browser and sign in (or use the free upload if it is still available on your account).
2. Upload a sharp phone photo of the bottle on a plain table (see section 4). Do not upload a photo with another product behind it.
3. Download the transparent PNG.
4. Open Canva (free) or any editor: set a square 2000 × 2000 canvas, fill with pure white, place the PNG, size the bottle to about 90% height, centre it, export PNG or JPG.
5. Check corners for leftover shadows. Hold the real bottle next to the screen: same blue, same logo, same cap.

Expected output (yours will be similar, not the same): a white-background listing photo that still looks like your real bottle — marker for health check: EXPANDOCT23-REMOVEBG-WHITE-MEESHO.

11. Worked example 5: Flipkart lifestyle scene with Photoroom

Suppose the second listing image should show the bottle on a kitchen shelf (Flipkart lifestyle slot). Photoroom is popular in India for product scenes. Do not invent fake brand awards on the bottle.

Steps:

1. Open Photoroom (app or web) and upload the same bottle photo.
2. Remove the background, then pick a kitchen / shelf template that looks real (wood shelf, soft daylight).
3. In the edit prompt or text field, write only scene words — not product redesign:

Keep this exact stainless steel bottle and blue cap.
Place it upright on a wooden kitchen shelf with soft morning light.
No new logos, no price tags, no extra bottles, no watermark.

4. Export a tall or square image that matches Flipkart's current size guide for that slot (check seller help — sizes change).
5. Compare with the real bottle again before upload.

Expected output (yours will be similar, not the same): the same bottle on a clean kitchen shelf, ready as a secondary listing image.

12. Worked example 6: Meesho 5-image set with Pixelcut resize

Suppose we need five Meesho images: main white, lifestyle, detail of the cap, size in hand, and packing. Pixelcut helps with batch background remove and resize. Free limits and paid plans change — check Pixelcut before you promise a client a deadline.

Steps:

1. Shoot five phone photos of the same bottle (main, lifestyle, close-up, in-hand, packing) under the same light.
2. In Pixelcut, use batch background remove on the main and close-up shots; place on white.
3. Use resize / export presets aimed at marketplace square sizes (often 1080 or 2000 px — confirm Meesho's current rule).
4. Name files clearly: meesho-bottle-01-main.jpg … 05-pack.jpg.
5. Upload in order. Never mix another SKU's photos into this set.

Expected output (yours will be similar, not the same): a five-image Meesho set with matching colour and no watermark on the main image.

13. Beginner mistakes and quick fixes

14. Try at home

Recap

Take one product photo on a white sheet near a window today and run the check script. Fix the start photo first, and every AI image after it gets better.

Previous chapter: ChatGPT Image Generation: Posters, Ghibli-Style Art and Thumbnails

Next chapter: Image-to-Video Explained: Direct a Short AI Clip from One Photo

Ravindra Bagale, trainer: linkedin.com/in/ravindra-bagale. The business, product, sizes and test images on this page are made up for learning. Amazon's main image rules were checked on Amazon seller help pages and style guides on 23 October 2026 and can change; always read the current rules for your category on each marketplace. The Python output was produced on 23 October 2026 with Pillow. AI images change every time, so your output will be similar, not the same.