I made 15 product videos in two days and learned that lighting matters more than motion
I sell home fragrance products on TikTok Shop. Soy candles, reed diffusers, wax melts. I source from a supplier in New Jersey, package them with my own labels, and ship from a small warehouse space I share with another seller in Brooklyn. My product photos are decent—I shoot them myself with a ring light—but I'd never made videos. Every TikTok Shop guide says video listings convert better than static images, but hiring a creator costs $150–$300 per video, and I have 40 SKUs.
Last week I decided to test AI video tools. I used Luma Ray 2 and Nova Reel on Kryotta to generate 15 product videos: five close-up spins, five lifestyle scenes, and five with text overlays. I uploaded them as TikTok Shop listing videos and tracked what happened over nine days. I looked at view-through rate (how many people watched past three seconds), cart adds, and actual purchases.
Four videos worked. They got cart adds and at least one sale each. Eleven were useless. People scrolled past in under two seconds, or they watched but didn't click. One video got comments asking if the product was real because the motion looked "off."
Here's what I did, what failed, and what actually converted.
The brief I gave each model
I started with product photos I already had: a vanilla bean candle in a white ceramic jar, a set of three wax melts in kraft paper packaging, a glass reed diffuser with rattan sticks. I uploaded each photo to Kryotta and wrote prompts for two different tools.
For Luma Ray 2, I kept prompts short and focused on one motion: "Slow 360-degree rotation of the candle on a wooden surface, soft natural light from the left, camera stays fixed." I wanted smooth spins with no weird warping.
For Nova Reel, I tried scene prompts: "Candle sitting on a coffee table in a cozy living room, morning light through sheer curtains, slight camera push-in." I wanted lifestyle context without having to stage a room.
I generated three versions of each video and picked the best one. Total cost was around $18 for all 15 videos. Each video was five seconds long, which TikTok Shop recommends for listing videos.
The close-up spins: two worked, three looked like melting wax
I made five 360-degree product spins. These are supposed to show the product from every angle, like the spins you see on Apple's website.
Two of them worked. The vanilla candle spin and the wax melt packaging spin both looked clean. The rotation was smooth, the lighting stayed consistent, and the product didn't warp. I uploaded them as listing videos. The vanilla candle video got 340 views in the first three days and added 12 items to cart. Two of those converted to sales. The wax melt video got 180 views and four cart adds, one sale.
Three of the spins failed. The reed diffuser spin had a moment halfway through where the glass bottle seemed to bend inward, like it was made of rubber. It looked wrong. I got one comment: "Is this a real product or AI?" The lavender candle spin had a lighting problem—the shadows rotated in the opposite direction from the candle, which made it look like the candle was spinning in place while the world moved around it. The third fail was a tin candle where the label text blurred and then re-sharpened as it rotated. It looked like a rendering glitch.
I didn't upload any of the three bad spins. If the motion looks uncanny, people assume the product is fake or low-quality.
The lifestyle scenes: one was perfect, four were uncanny
I tried five lifestyle scene videos using Nova Reel. The idea was to show the product in a real environment—on a nightstand, on a bathroom counter, on a bookshelf—so people could imagine it in their own space.
One video worked beautifully. It was the reed diffuser on a wooden side table next to a stack of books. The camera did a slow push-in, the lighting was warm and even, and nothing moved except a very slight shadow shift. It felt like a real room. That video got 290 views, nine cart adds, and two sales. It outperformed most of my static images.
The other four were disasters. One had a candle on a nightstand, but the nightstand had a drawer that seemed to phase in and out of existence as the camera moved. Another had a wax melt on a bathroom counter with a mirror in the background, and the reflection didn't match the product angle. It was subtle, but once you noticed it, you couldn't unsee it. The third had a candle on a bookshelf where the book spines kept changing. The fourth had lighting that shifted from warm to cool halfway through the video, like someone flipped a switch.
Uncanny motion is worse than no motion. If something feels wrong, people leave.
The text overlay videos: one converted, four got skipped
I made five videos with text overlays. These were static product shots with animated text on top: "Hand-poured soy wax," "Burns for 40 hours," "Free shipping over $35." I used Luma Ray 2 to add a slow zoom on the product while the text faded in.
One of these worked. It was the vanilla candle with the text "Smells like a bakery." Simple, direct, and the zoom was subtle enough that it didn't distract from the text. That video got 220 views, six cart adds, and one sale.
The other four got scrolled past. The problem wasn't the text—it was the pacing. Three of the videos had the text appear too fast (under one second) and then sit there for the remaining four seconds. People read the text and left. One video had the text fade in so slowly that it wasn't fully visible until second three, which is when most people decide to scroll.
TikTok Shop listing videos need to deliver information in the first second. If your text takes two seconds to appear, you've already lost half your viewers.
What I learned: start with the simplest motion that works
The videos that converted had three things in common. First, the motion was minimal. A slow rotation or a gentle push-in, nothing fast or complex. Second, the lighting was consistent. No shadows that moved wrong, no color shifts. Third, the product stayed in focus and didn't warp.
The videos that failed tried to do too much. Complex camera moves, multiple objects in the scene, reflections, fast text animations. AI video tools are good at simple, controlled motion. They're not good at physics, reflections, or anything with multiple moving parts.
If you're making product videos for TikTok Shop, start with a 360-degree spin on a plain surface. Use your existing product photos as the starting frame. Write a short prompt: "Slow rotation, soft even lighting, no background distractions." Generate three versions and pick the one where nothing looks wrong. If the spin works, try a simple zoom or a slow push-in. If that works, then try a lifestyle scene with one product and one surface.
Don't start with the lifestyle scene. You'll waste time and money on videos that look almost right but feel wrong.
I'm still using my ring light and my phone for hero images, but now I'm adding AI video to about half my listings. The four videos that worked are still live, and they're converting better than static images. I'm not replacing my product photography—I'm adding motion where it helps. I'll make another batch of 10 videos next week, all simple spins, and see if I can get the success rate above 50%.
I use Kryotta because I can test both Luma and Nova Reel without switching tools, and the side-by-side preview makes it easy to spot bad motion before I upload anything. If you sell physical products and you've been putting off video because it's expensive or complicated, try five simple spins first. You'll know in a week whether it's worth doing more.
Questions people ask
How long should a TikTok Shop product video be?
Five to seven seconds. TikTok Shop recommends under 10 seconds for listing videos. Most people decide to scroll or click within three seconds, so front-load the most important visual information.
Can I use AI video for the main listing video or just as extra content?
You can use it as the main video, but I'd keep a high-quality static photo as the first image in your listing gallery. Some buyers still prefer photos, and if your video has any uncanny motion, the photo gives them a fallback.
Which tool is better for product videos, Luma or Nova Reel?
Luma Ray 2 is better for controlled motion like spins and zooms. Nova Reel is better for adding scene context, but it's less reliable—you'll generate more duds. Start with Luma for your first batch.
Do I need to disclose that the video is AI-generated?
TikTok Shop doesn't require it for product listing videos, but if the product itself looks different from the video (wrong color, wrong size, wrong texture), you'll get returns. Use AI video to show motion and context, not to misrepresent the product.
If you want to test AI video without committing to a monthly subscription or a freelancer, you can try both tools and compare outputs side by side at Kryotta. Generate a few spins, see what works, and go from there.
