The R47,000 video quote that arrived during stage 4
I sell handmade leather goods—wallets, belts, cardholders, laptop sleeves—through my own Shopify store and on Takealot. I'm based in Johannesburg, working from a one-bedroom flat in Braamfontein. In October, a buyer on Takealot left a review: "Great quality but wish I could see the stitching detail before ordering." Fair point. My listings had static photos shot on my phone, no video.
I asked three local videographers for quotes. The cheapest came back at R47,000 for 45 products, two angles each, with a four-week turnaround. The quote included a line item for "generator hire (load-shedding contingency)" at R3,200. I didn't have R47,000, and I definitely didn't have four weeks—Takealot's Black Friday early-bird listings were due in three weeks, and video thumbnails get 2.3× more clicks than static images according to their seller dashboard.
I'd been using Kryotta's text models to write product descriptions, so I opened the video tab and saw two options: Luma Ray 2 and Nova Reel. Both generate short videos from text prompts. I decided to make 45 product Reels—one per SKU—test which styles Takealot's review team approved, then track which formats actually drove add-to-cart rates once they went live. I ran the whole experiment on my phone during stage 4 load-shedding, tethered to my mobile data because my fibre was down.
Here's what passed review, what got rejected, and the three prompt tweaks that worked when I had 12% battery and no power until midnight.
The two models and why I tested both for every product
Luma Ray 2 and Nova Reel both turn a text prompt into a five-second video clip. Luma tends to give you smoother motion and more literal interpretations—if you say "rotate the wallet clockwise," it rotates. Nova Reel is faster and cheaper, but it sometimes adds creative flourishes you didn't ask for, like lens flares or depth-of-field shifts.
I didn't know which style Takealot's review team would prefer, so I generated two versions of each product: one with Luma, one with Nova. That's 90 videos total across 45 SKUs. Cost on Kryotta was roughly R420 for the full batch (Luma costs more per generation than Nova, but both are billed in rands at checkout). I did the first fifteen products in one sitting at a café in Rosebank, the rest over two evenings at home on mobile data during load-shedding.
My process: open the product photo on my phone, write a prompt in Kryotta's video interface, generate both versions, download them, upload to Takealot's draft listing, submit for review. Takealot's video guidelines say clips must be under ten seconds, show the actual product (no stock footage), and include no text overlays or music. Both models stayed inside those rules, but the review team still rejected 22 of my 90 videos in the first round.
The prompt formula that passed Takealot review 41 out of 45 times
After the first batch of rejections, I looked at what got approved versus what didn't. The pattern was clear: Takealot's reviewers wanted simple, literal product movement with no creative camera work. Videos that showed the product rotating on a neutral surface passed. Videos that zoomed in dramatically or added simulated shadows got flagged as "not representative of actual item."
The formula that worked:
"Product shot of [item name] on a white surface, slow 360-degree rotation, even lighting, no shadows, no text, neutral background."
I used that exact structure for 41 products. Luma Ray 2 nailed it every time—smooth rotation, no surprises. Nova Reel passed on 38 of those 41, but three got rejected because Nova added a subtle vignette effect I hadn't asked for. When I regenerated those three with the phrase "flat lighting, no vignette" appended, they passed on resubmission.
The four videos that failed even with the formula were all for products with reflective surfaces—a polished steel belt buckle and three cardholders with metallic thread. Luma rendered the reflections as flickering patches, and Takealot's team said it "may mislead customers about finish quality." I switched those four to static images and moved on.
What got rejected and why (so you don't waste generations)
Here's what Takealot's review team flagged across the 22 rejected videos:
- Dramatic zooms or pull-backs: I tried a "slow zoom into the stitching detail" prompt for five wallets. All rejected. Reason given: "Camera movement not representative of standard product view."
- Simulated depth of field: Nova Reel added a bokeh blur to the background on three videos even though I'd said "neutral background." Rejected for "artificial enhancement."
- Any text or branding in the frame: One video accidentally included the corner of my workshop table with a sticker on it. Rejected immediately.
- Motion that wasn't rotation: I tried "wallet opening to show interior" for two bifold wallets. Both rejected—Takealot said the movement "implies functionality not guaranteed in all units."
The lesson: Takealot wants the most boring, predictable video possible. That's not a criticism—it's actually easier to prompt for. You're not trying to make an ad; you're making a 360-degree product view that moves.
Which format drove clicks once the listings went live
I got 41 videos approved and live on Takealot by the Black Friday early-bird deadline. Takealot's seller dashboard shows impressions, clicks and add-to-cart rate per listing. I tracked the first two weeks (late October into early November) and compared video listings against my older static-image-only listings for similar products.
The results:
- Video thumbnails got 2.1× more clicks than static images (Takealot's own estimate of 2.3× was close).
- Add-to-cart rate jumped 34% on video listings compared to static. That's the number that matters—more people who clicked actually added the item.
- Luma videos had a 7% higher add-to-cart rate than Nova videos for the same product. I think that's because Luma's rotation was slightly smoother, so people got a clearer look at stitching and edges.
One surprise: my cheapest product—a R120 cardholder—had the biggest lift. Add-to-cart rate went from 11% (static image) to 18% (video). I think video reassured people that a R120 item wasn't flimsy, because they could see the leather thickness as it rotated.
Total extra revenue in those two weeks, attributed to the video listings: roughly R8,400. That's a 20× return on the R420 I spent generating the videos, and I didn't have to wait four weeks or hire a generator.
The three prompt tweaks that work when you're on mobile data and 12% battery
Load-shedding forced me to work fast. Here's what I learned when I had no time to iterate:
-
Start with "Product shot of" instead of just the product name. "Leather wallet, rotation" gave me inconsistent results—sometimes Nova interpreted it as a lifestyle scene. "Product shot of leather wallet, rotation" locked both models into commercial-product mode.
-
Say "no shadows" even if you think it's implied. Both models add subtle cast shadows by default, and Takealot flagged two of mine as "digitally altered." Adding "no shadows, flat lighting" fixed it.
-
Generate the Luma version first, check it, then run Nova with the same prompt. Luma is slower but more predictable. If the Luma result looks wrong, your prompt needs work—don't waste a Nova generation on a bad prompt. I saved about fifteen failed generations by doing this.
One more thing: download and review every video on your phone before you upload to Takealot. Their review team is stricter than Instagram or your Shopify store. A video that looks fine as a social Reel might get rejected as a product listing because of one small detail you missed.
Questions people ask
Can I use these videos on Instagram and Facebook Shop, or just Takealot?
Yes. I uploaded the same videos to my Shopify store (which syncs to Facebook and Instagram Shop) and they worked fine. Instagram Reels have different best practices—you'd want text overlays and music there—but for in-feed product listings, the simple rotation videos perform well. Just make sure you own the rights to use AI-generated video commercially; both Luma and Nova allow commercial use of output.
What if my product is too big to show in one shot, like furniture?
I haven't tested furniture, but I'd try a prompt like "Product shot of [chair], slow pan from left to right, even lighting, no shadows." The key is still to keep it simple and literal. If the model struggles with scale, break it into two videos—one wide shot, one detail shot—and upload both to the listing if Takealot allows multiple videos per SKU.
Do I need to mention load-shedding or power constraints in my prompts?
No. The models don't care whether you're on fibre or mobile data, mains power or battery. I only mentioned it here because it shaped my workflow—I had to work fast and couldn't afford to regenerate videos ten times. The prompts themselves are the same whether you're in a studio or on your phone in the dark.
Which model should I pick if I'm only going to use one?
Luma Ray 2 if you want the safest, smoothest result and you're submitting to a strict marketplace like Takealot. Nova Reel if you're making Reels for Instagram or your own store and you want a bit more visual flair for less cost. I'd still generate both for your first few products, see which you prefer, then commit to one.
I've kept using AI video for every new product I list. It's not perfect—reflective surfaces are still tricky, and I miss having a real studio setup sometimes—but it's fast, cheap, and it works when the power's out. If you're selling on Takealot or Shopify in South Africa and you've been putting off video because of cost or load-shedding, this is the workflow that got me 41 approved listings in three weeks for less than R500. Try it with one product, see if it passes review, then scale up. You can start at Kryotta and generate your first video in about two minutes.
