I made 50 product Reels for Instagram Shop with Luma and Nova: which style got saves

I ran the test during Diwali prep—and learned saves matter more than views

I run a Shopify store selling brass diyas, marble coasters, hand-painted rangoli stencils and gift hampers. Most of my orders come from tier-2 cities—Jaipur, Coimbatore, Indore—and about 60 per cent of my traffic lands on Instagram Shop before clicking through to the site. Last October I had three weeks to build a Reel library before Diwali orders peaked, and I didn't have the budget to hire a product videographer at ₹2,000 per shoot.

I decided to generate 50 Reels using three AI video tools: Luma Ray 2, Nova Reel, and Stability's image-to-video model (via SD 3.5). I already had clean product photos from my last catalogue shoot, so I fed those in as starting frames and tested different animation styles, prompt patterns and music choices. Then I posted them over three weeks, tagged each one in a spreadsheet, and tracked saves, shares, two-second drop-off rate and cart adds via Instagram Insights and Shopify UTM links.

The results surprised me. The Reels that got the most views weren't the ones that drove saves or cart adds. And the model I thought would be cheapest ended up costing more per conversion because I had to regenerate so many duds. Here's what worked, what didn't, and which prompts I'm using for wedding season.

Luma Ray 2 won on saves when I prompted for "slow reveal" camera moves

Luma gave me the highest save rate: 4.2 per cent across 18 Reels. The ones that performed best used a "slow push-in" or "gentle rotate" camera move with no cuts. My best prompt was "Slow push-in on brass diya set on marble surface, soft golden hour light, camera moves from wide to close-up over 5 seconds, no cuts, product stays centred". That Reel got 140 saves and 22 cart adds from 3,300 views.

The reason it worked: people saved it as a gift idea or home-décor reference. The slow movement let them actually see the product detail—the engraving on the diyas, the way the light caught the brass. Fast cuts or flashy transitions tanked the save rate because viewers couldn't screenshot or remember what they'd seen.

Luma's weakness: it occasionally invented details that weren't in my photo. One Reel added a candle flame I hadn't lit, another changed the colour of a rangoli stencil from red to orange. I caught those before posting, but I had to regenerate about 30 per cent of outputs. That pushed my effective cost to around ₹18 per usable Reel, assuming I value my time at zero.

Nova Reel was fastest and cheapest—but only for static "floating product" shots

Nova Reel gave me the lowest cost per Reel: about ₹8 each, and most came out usable on the first try. The style that worked best was a simple "product floating on clean background, gentle rotation, no camera move". Prompt: "Marble coaster set rotating slowly on white background, soft studio lighting, 4 seconds, clean and minimal".

These Reels didn't get high save rates—around 2.1 per cent—but they had the lowest two-second drop-off (18 per cent vs 31 per cent for my Luma Reels). People watched them all the way through, and the cart-add rate was decent: 1.8 per cent, which is better than my static-image posts usually manage.

Nova's trade-off: it doesn't handle complex scenes. When I tried a prompt with multiple products or a textured background (a diya set on a silk cloth), it blurred the edges or made the cloth look like melted plastic. If your product photos are already clean and isolated, Nova is fast and cheap. If they're lifestyle shots with props, you'll fight it.

Stability image-to-video added motion to the background, not the product—and viewers hated it

I generated 14 Reels with Stability's image-to-video model, and only two got a save rate above 1 per cent. The problem: it animated everything except the product. My brass diyas stayed still while the marble surface rippled like water, or the background curtain swayed as if a fan were blowing. One Reel made my rangoli stencil look like it was sliding off the table.

The only Stability Reels that worked used a prompt that locked the product in place and animated only the light: "Brass diya on stone surface, product stays completely still, only the candlelight flickers gently". That got 89 saves from 2,100 views, but I had to regenerate it six times to stop the diya itself from morphing.

I've stopped using Stability for product Reels. It's better suited to abstract B-roll or background loops, not hero shots where the product needs to stay recognisable.

The prompt pattern that lifted saves across all three models

Regardless of which model I used, one prompt structure consistently beat the others:

1. Lock the product in place. "Product stays centred and still" or "camera moves, product does not".
2. Specify one light change. "Soft golden hour light fades to warm evening glow" or "single spotlight brightens gradually".
3. Name the surface and background. "White marble surface, plain cream wall behind" instead of "elegant setting".
4. Set the duration. "4 seconds" or "5 seconds"—anything longer and the two-second drop-off spiked.

My worst-performing Reels used vague prompts like "Make this diya look festive and premium". The models added spinning text, sparkle overlays, or zoomed the camera in and out like a TikTok transition. Saves dropped to 0.6 per cent and cart adds disappeared.

Music choice mattered more than I expected—and trending audio didn't help

I tested three music strategies: trending Reels audio, royalty-free instrumental tracks, and silence with text captions. Trending audio gave me higher reach (Instagram pushed those Reels harder), but the save rate was lower—1.9 per cent vs 3.4 per cent for instrumental tracks.

The reason: people who saved my Reels were looking for gift ideas or décor inspiration, not entertainment. Trending Bollywood remixes or viral comedy audio made the product feel less serious. A simple instrumental sitar loop or a soft tabla track kept the focus on the product and lifted saves by 40 per cent.

Silence with captions worked for hampers and gift sets, especially when I added text like "Under ₹800 | Free shipping | COD available". Those Reels got fewer views but a higher cart-add rate: 2.3 per cent vs 1.6 per cent for the same product with music.

I tracked cart adds with UTM links in my Instagram Shop catalogue

Instagram Insights tells you saves and shares, but it doesn't tell you which Reel drove a sale. I added a unique UTM parameter to every product link in my Instagram Shop catalogue—utm_source=instagram&utm_medium=reel&utm_campaign=diwali_diya_luma_01—and checked Shopify's traffic sources report every morning.

Luma Reels drove 34 per cent of my Instagram cart adds during the test period, even though they made up only 36 per cent of my Reel posts. Nova Reels drove 28 per cent, and Stability drove 11 per cent. The rest came from static posts or Stories. The Luma Reels that got saved also got shared more—people sent them to family WhatsApp groups with messages like "This for Diwali?" which I could see in my DM requests.

One surprise: Reels posted between 7 PM and 9 PM got 60 per cent more saves than the same Reel posted at 11 AM. I assume people browse gift ideas in the evening after work, not during lunch.

What I'm doing differently for wedding season

I'm sticking with Luma for hero products—anything I want people to save or share as a gift idea. I'll use Nova for catalogue Reels where I just need a clean, quick loop to show the product exists. I've dropped Stability entirely for product work.

My new prompt template for Luma: "Slow 4-second push-in on [product] on [surface], [lighting], camera moves smoothly from wide to close-up, product stays perfectly still and centred, no text or effects". I regenerate anything that adds details I didn't photograph.

For music, I'm using instrumental tracks from YouTube Audio Library or Epidemic Sound (₹500/month) and skipping trending audio unless I'm promoting a flash sale where reach matters more than saves.

I'm also testing a hybrid: generate the Reel with Luma or Nova, then add a text overlay in CapCut with price, shipping and COD details. Early results show that lifts cart adds by about 15 per cent without hurting the save rate.

I built all 50 Reels in Kryotta and compared outputs side by side

I used Kryotta's compare arena to run the same prompt through Luma Ray 2, Nova Reel and Stability image-to-video at the same time. That let me pick the best output without switching tabs or re-uploading my product photo three times. For the Reels that worked, I saved the prompt and the model choice in a workspace folder so I can reuse the pattern for wedding-season products—silver trays, gift boxes, decorative bells.

If you're ramping up for wedding season or planning a Republic Day sale, test a few models with the same product photo before you commit to generating 50 Reels with one tool. The model that works for a brass diya might fail on a fabric hamper, and you won't know until you compare them. You can try the compare arena and video tools at Kryotta and see which model fits your catalogue.

Questions people ask

Which AI video model is cheapest for product Reels?
Nova Reel gave me the lowest cost per usable Reel—around ₹8—and worked well for simple floating-product shots on clean backgrounds. Luma cost more (around ₹18 per Reel after regenerations) but delivered higher save rates and cart adds for hero products.

Do AI-generated Reels hurt your Instagram reach?
Not in my test. My AI Reels got the same average reach as my photographer-shot Reels (around 2,800 views per post), and Instagram didn't flag or suppress any of them. The algorithm cares more about watch time and saves than whether the video is AI-generated.

Should I use trending audio or instrumental music for product Reels?
Instrumental tracks lifted my save rate by 40 per cent compared to trending Bollywood remixes. Trending audio gave me higher reach but lower saves and cart adds. Use instrumental for gift-idea content and trending audio only for flash sales where you need reach fast.

How do I stop AI video models from changing my product's colour or details?
Add "product stays exactly as shown in the image, no added details or colour changes" to your prompt, and regenerate anything that invents new elements. Luma was the worst offender for this; Nova stayed closer to the original photo.

I'm still learning which styles work for different product categories, but the save rate and cart-add data from this test has already paid for the three weeks I spent generating Reels instead of shooting them myself. If you're running an Instagram Shop and you're tired of static product photos, try generating five Reels with different models and see which one your audience saves. The model that wins might surprise you.