I made 20 product shots for Amazon and Flipkart with AI: Stability vs Ideogram vs Midjourney

The three products I chose and why they're hard to photograph

I picked three items that show up constantly in my WhatsApp seller groups: a Bluetooth speaker (electronics), a cotton kurta (clothing), and a ceramic planter (home decor). All three are nightmare categories for DIY photography. The speaker's glossy black plastic shows every fingerprint and reflection. The kurta is pale yellow—shoot it wrong and it looks either washed-out or jaundiced on a phone screen. The planter has a matte glaze that photographs flat unless you get the lighting exactly right.

Professional product shoots in Bangalore or Mumbai start around ₹800 per SKU for white-background images. If you're launching twenty products, that's ₹16,000 before you've sold a single unit. Most tier-2 sellers I know—myself included—end up shooting on their dining table with a bedsheet backdrop and praying the image clears Amazon's quality check.

I wanted to see if AI image generators could do better, and whether the cost made sense for someone managing listings on Amazon.in and Flipkart without a studio budget.

The tools I tested and what they cost in rupees

I tested three platforms over one week:

Stability AI (via Kryotta): Stable Image Core and SD 3.5 Large. Kryotta includes Stability models in the workspace, so I was already paying for access—no separate subscription. Roughly ₹12 per image at standard resolution.

Ideogram: Their v2 model, which several seller groups recommended for product shots. Pricing is $10 for 400 credits (around ₹840), and a high-quality image costs 20 credits—so about ₹42 per image.

Midjourney: v6.1 on the $10 Basic plan (around ₹840/month), which gives you roughly 200 fast generations. Works out to ₹4 per image if you use the full quota, but you're paying the monthly fee whether you generate twenty images or two hundred.

I generated five versions of each product with each tool—fifteen images per tool, sixty images total once I included variations and retries. Then I ran the best candidates through Amazon.in's image upload checker to see what passed.

What worked: the Bluetooth speaker on Stability

The speaker was the easiest. I used Stable Image Core in Kryotta with this prompt:

"Product photography of a black Bluetooth speaker on pure white background, centered, professional e-commerce lighting, sharp focus, no shadows, studio quality, front view"

First attempt: usable. The speaker looked real, the background was clean white (RGB 255,255,255—Amazon checks), and the edges were sharp. I uploaded it to Amazon Seller Central and it cleared the automated quality check immediately. No red flags for background colour, no warnings about resolution.

I tried SD 3.5 Large with the same prompt and got an even better result—slightly more detail in the speaker grille, better highlight on the glossy top. The model handled reflections well, which is where phone photos usually fail. Cost: ₹12 for the Core version, maybe ₹18 for Large. Both cheaper than reshooting with a lightbox.

The Ideogram version looked good but had a faint grey gradient in the background—not enough to see on my laptop, but Amazon's validator flagged it. I could have fixed it in an editor, but that defeats the point of a one-click solution.

Midjourney gave me a gorgeous image with dramatic side lighting and a subtle shadow. Beautiful for Instagram, useless for Amazon. Their model seems trained on aspirational product photography, not the flat, boring, compliant style that marketplaces want. I tried adding "no shadows, pure white background" to the prompt and got a speaker floating in a void with a glow effect. Artistic, but not what I needed.

Stability won the electronics category. Fast, cheap, compliant.

What failed: the cotton kurta on all three platforms

Clothing is where all three tools struggled. I started with this prompt:

"Product photography of a pale yellow cotton kurta on pure white background, front view, flat lay, no model, professional e-commerce lighting"

Stability Core gave me a kurta with three sleeves. Not a typo—three distinct sleeves, one of them emerging from the side seam. I regenerated four times and got: two sleeves but the neckline was inside-out, a kurta with no visible stitching (it looked vacuum-formed), one decent result but with a beige background, and one that was perfect except the fabric texture looked like vinyl.

SD 3.5 Large did better—two proper sleeves, correct neckline, reasonable cotton texture—but the colour shifted to mustard yellow under Amazon's white-balance check. Flipkart accepted it, Amazon didn't.

Ideogram produced a kurta that looked hand-drawn. The fabric folds were too uniform, like a 3D render from 2015. It might work for a mood board, but I wouldn't list it on Meesho.

Midjourney gave me a beautiful lifestyle shot: kurta on a wooden hanger, brick wall background, morning light. Completely wrong for a listing photo, but I saved it for Instagram.

None of these tools are ready for clothing listings yet. The fabric physics are off, the colours drift, and the details—buttons, stitching, hems—go wrong more often than right. I'd still shoot clothing on a mannequin or pay for a proper studio day.

What surprised me: the ceramic planter on Ideogram

The planter was the surprise winner for Ideogram. I used this prompt:

"Product photography of a small ceramic planter, matte grey glaze, pure white background, centered, top-down angle, professional e-commerce lighting"

Ideogram v2 nailed it on the second try. The glaze looked tactile, the rim had realistic thickness, the drainage hole was visible and correctly placed. I uploaded it to both Amazon and Flipkart and it passed both validators. Cost: ₹42.

Stability Core gave me a planter with a wavy rim—like it had melted slightly—and SD 3.5 Large produced something that looked more like concrete than ceramic. Both were close, but the texture was wrong.

Midjourney gave me a planter on a marble countertop with a succulent already planted and a linen napkin artfully draped beside it. Again, gorgeous, again useless for a white-background listing.

Ideogram surprised me for home decor. It's pricier than Stability, but if you're listing planters, candles, cushions—anything with texture—it's worth testing.

The compliance failures I didn't expect

Amazon.in's image validator rejected six of my images for reasons I didn't anticipate:

• Shadow under the speaker (Midjourney): even a faint drop shadow counts as non-white background.
• Colour cast (Stability, kurta): the yellow shifted warm and failed the white-balance check.
• Visible pixels (Ideogram, planter, first attempt): the resolution was fine, but the validator flagged "low quality" because of compression artifacts. I regenerated at higher quality and it passed.
• Floating product (Midjourney, speaker): the dramatic lighting made the speaker look like it was hovering. Amazon wants products grounded.
• Text in the image (my mistake): I tried adding "New Arrival" to one planter shot and forgot AI models sometimes render text as gibberish. Amazon rejected it for "prohibited text."

Flipkart's validator is less strict—it accepted two images Amazon rejected—but I wouldn't rely on that. If you're cross-listing, aim for Amazon's standard.

The actual cost for twenty products

Here's what it would cost to generate one compliant image per product for twenty SKUs, assuming you get it right on the second attempt (I didn't, but let's be optimistic):

• Stability via Kryotta: 20 images × ₹12 = ₹240. Add another ₹120 if you need retries. Total: ₹360.
• Ideogram: 20 images × ₹42 = ₹840. Retries push it past ₹1,000.
• Midjourney: ₹840/month whether you use it or not. If you're only generating product shots, you're paying for a tool that's built for something else.

For comparison, a basic studio shoot in a tier-2 city costs ₹500–800 per product. Twenty products: ₹10,000–16,000.

Stability is the budget winner if you're listing electronics or home decor and you don't need perfection. Ideogram is worth the extra cost for textured items. Midjourney is beautiful but wrong for marketplace listings—save it for social content.

The workflow I'm using now

I don't use AI for every product photo. But here's the checklist I follow when I do:

1. Electronics and hard goods: Stability via Kryotta. Prompt: "Product photography, pure white background, centered, front view, professional e-commerce lighting, no shadows." Generate two, pick the better one, upload.

2. Home decor with texture: Ideogram v2, same prompt structure. Budget ₹80 per product (two attempts).

3. Clothing: still shooting on a mannequin. AI isn't reliable enough yet.

4. Lifestyle shots for Instagram: Midjourney. The dramatic lighting and styled backgrounds work better for social than for listings.

5. Always check the validator: upload to Amazon Seller Central before you commit to an image. The automated checker catches things your eye won't.

I'm saving about ₹8,000 per product launch compared to studio shoots, and the images clear compliance checks roughly 70% of the time on the first try. That's good enough for me to keep using AI for half my catalogue.

Questions people ask

Can I use AI images on Amazon.in without disclosing they're AI-generated?
Yes. Amazon's image guidelines don't require disclosure, as long as the image accurately represents the product you're selling. Don't add AI-generated features the real product doesn't have.

Which tool is best for white-background product photos in India?
Stability AI via Kryotta for cost and compliance. It's built into the workspace, it's cheap per image, and it handles white backgrounds reliably for electronics and decor. Ideogram is better for texture but costs more.

Do AI product images hurt conversion rates compared to real photos?
I haven't seen a drop. Customers care whether the image is clear, accurate and loads fast. If your AI image is sharper than your phone photo, it'll probably perform better.

Can I generate images of clothing that pass Amazon's quality check?
Not reliably yet. Fabric folds, stitching and colour accuracy are still problems. I'd only use AI for clothing if you're willing to regenerate five times and accept a 50% success rate.

I'm not ditching my phone camera entirely, but for electronics and home decor I'm generating first and shooting only if the AI version fails. It's faster, cheaper, and honestly less stressful than trying to get the lighting right on my dining table at 11 p.m. before a listing deadline.

If you're managing your own Amazon or Flipkart store and you're tired of paying for studio shoots, try Stability in Kryotta for your next product launch—it's already included, and you'll know in two minutes whether it works for your catalogue.