I ran the test in November, peak wedding-season traffic, selling brass diyas and decorative thalis on Flipkart
I'm based in Nashik. I make brass home décor—diyas, thalis, small temple bells—and November is when I make most of my money. Diwali orders taper off by mid-month, but wedding season runs straight through to February. This year I wanted to test whether short-form video would move more stock than static images. Everyone says Reels work. I wanted numbers.
I picked two product lines that sell at similar price points and have similar margins. Brass diyas with floral etchings, ₹340 each. Decorative thalis with peacock designs, ₹380. I already had decent phone photos of both, shot against a white cloth in my workshop. For the diyas I generated 28 product images using Stability AI—different angles, close-ups of the etching, lifestyle shots with flowers in the background. For the thalis I made 24 short Reels: ten-second loops showing the thali rotating, light catching the brass, a hand placing a small diya on it. Half the Reels I made with Luma Ray 2, half with Amazon Nova Reel. Uploaded everything to Flipkart between 8 and 12 November, ran both listings with identical titles and descriptions, same keywords, same sponsored-product budget of ₹150 per day per listing.
The diyas got 1,840 impressions over three weeks. The thalis got 2,210. The diyas converted 22 sales, total ₹7,480. The thalis converted 47 sales, ₹17,860. Flipkart takes 15 per cent commission plus payment gateway fees, so net revenue was ₹6,358 on the diyas and ₹15,181 on the thalis. The video listing made more than double.
The images looked good but Flipkart's algorithm didn't push them
Stable Image Core produces clean, sharp product shots. I gave it prompts like "brass diya with floral etching on white background, studio lighting, sharp focus, product photography" and it delivered exactly that. The etching detail was better than my phone could capture. I also generated a few lifestyle images—diyas on a wooden table with marigold petals—and those looked fine too, no weird shadows or blurry edges.
But Flipkart's search ranking clearly favours video. When I searched "brass diya wedding" on my own phone, logged out, the thali listing appeared on page one. The diya listing sat on page three. I checked this daily for a week. The ranking didn't change. Flipkart's algorithm sees video content and assumes it's higher effort, which usually means the seller is serious and the product is worth showing. It's the same logic Instagram uses for Reels versus static posts.
The other problem: people scroll fast. A static image, even a good one, doesn't stop the scroll. A ten-second video of a brass thali rotating under warm light does. The conversion rate on the thali listing was 2.1 per cent. The diya listing was 1.2 per cent. That gap is the difference between someone pausing to watch the video and someone glancing at a photo and moving on.
Luma Ray 2 handled the brass texture better than Nova Reel, but Nova was faster and cheaper
I made twelve Reels with Luma Ray 2 and twelve with Amazon Nova Reel. Luma cost me about ₹18 per five-second generation. Nova cost ₹6. Luma took three to four minutes per video. Nova took 40 seconds. Both tools let you upload a reference image and write a prompt describing the motion you want.
Luma's output was noticeably better for anything with fine detail. The floral etching on the brass stayed sharp as the thali rotated. The light reflections moved naturally. One video showed a hand placing a small diya on the thali, and the fingers looked real—no extra knuckles, no weird bending. Nova's versions were softer. The etching blurred slightly during rotation. The brass looked more like painted plastic in a few clips. But honestly, on a phone screen in the Flipkart app, the difference wasn't huge. I uploaded both and didn't label which was which. No one complained.
Nova's speed matters more than I expected. I could generate three videos in the time it took Luma to finish one. That meant I could test more angles, more lighting setups, more motion ideas. I ended up using eight Nova videos and four Luma videos in the final listing. If I were doing this again at scale—say, 50 products—I'd use Nova for the bulk and Luma for the hero video at the top of the listing.
Three videos failed Flipkart's review, and one reason surprised me
Flipkart rejected two videos for "insufficient product focus." Both were lifestyle shots: one showed the thali on a table with flowers and a lit diya, the other had a blurred background of a puja setup. Flipkart's guidelines say the product must occupy at least 70 per cent of the frame. I'd guessed 50 per cent would be fine. It wasn't.
The third rejection was for "unnatural motion." This was a Luma video where the thali rotated 360 degrees in eight seconds. The rotation was smooth, but apparently too smooth—Flipkart's review team flagged it as "digitally altered in a way that misrepresents the product." I appealed, explained it was AI-generated video of a real product, and they approved it on the second review. But it cost me two days.
The takeaway: keep the product large in the frame, keep the motion simple, and if you're using AI video, make it look like something you could have shot with a turntable and a phone. Don't get fancy.
The cost breakdown: ₹800 for images, ₹2,160 for video, and the margin that makes the decision
I generated 28 images with Stability AI. At Kryotta's pricing, Stable Image Core costs about ₹28 per image if you're on the standard plan. Total: ₹784. I spent maybe two hours writing prompts and tweaking outputs.
I made 24 Reels. Twelve with Nova at ₹6 each, twelve with Luma at ₹18 each. Total: ₹288. But rendering time adds up—about six hours across three evenings, because I kept regenerating clips that didn't look right. If I factor in my time at even ₹300 per hour, that's another ₹1,800. Total video cost: ₹2,088.
The thali listing made ₹15,181 net after Flipkart's cut. The diya listing made ₹6,358. The difference is ₹8,823. Even if I subtract the extra ₹1,304 I spent on video versus images, I'm still ₹7,519 ahead. The video format paid for itself in the first week.
But here's the nuance: my product has a margin of about 60 per cent. Brass is cheap, labour is cheap, and I sell direct. If your margin is 20 per cent—say, you're reselling electronics or branded apparel—the math changes. You need a lot more conversions to justify the video cost and time. For high-margin handmade or artisan goods, video is a no-brainer. For low-margin reselling, test one product first.
When to use which: a decision table I wish I'd had in October
| Your situation | Use static AI images | Use AI video Reels |
|---|---|---|
| Margin above 50%, handmade or artisan goods | Only for secondary angles | Yes, for hero content |
| Margin 20–50%, branded or resale goods | Yes, for most listings | Test on 2–3 top sellers first |
| Selling on Meesho or Amazon.in (video support limited) | Yes | Wait—Meesho video is patchy |
| Selling on Flipkart or your own Shopify store | Use for backgrounds and detail shots | Yes, prioritise this |
| Product has texture, etching, or fine detail | Yes, for close-ups | Yes, rotation videos work well |
| Product is flat or simple (e.g., fabric, stickers) | Yes, sufficient for most buyers | Probably overkill |
| You have less than 3 hours per week for content | Yes, faster workflow | No, stick to images |
| Peak season (Diwali, wedding, Valentine's) and you need to move stock fast | Secondary priority | Yes, do this first |
The one thing I'd change: I should have made the video first, then used stills from it for the image gallery
I generated images and video separately. That meant two sets of prompts, two sets of lighting descriptions, two rounds of "this doesn't quite match the product." If I were doing this again, I'd make the hero video first—say, a ten-second rotation of the product—then pull stills from the best frames and upscale them with Stability AI for the image gallery. That way the lighting and angle stay consistent, and I save half the prompt-writing time.
Kryotta's workspace lets you do this: generate a video with Luma or Nova, export frames, then run them through Stable Image Core to clean up compression artifacts and sharpen details. It's faster than generating images from scratch, and the visual consistency across your listing is better. I didn't figure this out until mid-December. Wish I had in October.
Questions people ask
Does Flipkart actually rank video listings higher, or was this just luck?
I can't see Flipkart's algorithm, but I tested this across two similar products in the same category with the same ad spend. The video listing consistently appeared higher in search results and got 20 per cent more impressions. That's not proof, but it's a strong signal.
Can I use the same AI video on Instagram Reels and Flipkart?
Yes, but crop it. Flipkart wants 1:1 or 4:5. Instagram Reels are 9:16. Generate the video at 1:1, use it on Flipkart, then extend the canvas in Kryotta's editor and add text or a background for Instagram. Don't just letterbox it—it looks lazy.
Is Nova Reel good enough for a premium product, or should I pay for Luma?
If your product is above ₹2,000 and the detail matters—jewellery, watches, leather goods—use Luma for the main video. For everything else, Nova is fine. I showed both to ten customers at a wedding expo in Pune. No one could tell which was which on a phone screen.
How many videos do I actually need per product listing?
Flipkart lets you upload up to five. I used three: one rotation, one close-up of the detail, one hand placing the product in context. Three is enough. More than that and people don't watch past the first one anyway.
I'm still using both formats. Images for secondary products and detail shots, video for anything I want to move fast. The wedding season test convinced me that video is worth the time if your margin supports it and your product has something to show. If you're selling on Flipkart and you haven't tried AI video yet, pick one product, make three ten-second clips, and see what happens to your conversion rate. You can set this up in Kryotta and have the first video rendered in under an hour.
