I had six days to make 30 product assets and no budget for a photographer
I sell handmade leather goods on Shopify—wallets, cardholders, tote bags, watch straps. Most of my customers come from Instagram and TikTok, where I post product shots and short Reels showing how a wallet ages over time or how the cardholder fits in a jacket pocket. Last October I knew Black Friday would be my biggest week of the year, but I had a problem: my product photography was stale. Same white background, same angle, same lighting I'd used since 2022. I needed fresh visuals, and I needed them fast.
I had six days before I wanted to start the pre-Black Friday email sequence. No budget for a photographer (I'd spent $800 on leather stock). No time to set up a lightbox and shoot 30 products myself—I tried that last year and it took four days for results that looked flat. So I tested AI. Stability AI for product images. Luma Ray 2 and Nova Reel for video ads. I made 30 assets: 18 product shots, 12 short videos. Some looked incredible. Some looked like a robot had drawn hands for the first time. I tracked which ones drove clicks during the sale, and I'm going to show you exactly what worked, what failed, and the checklist I'd use if I did this again.
The checklist: what to make, which tool to use, and what to watch for
Here's the process I followed, with notes on where things went wrong.
1. Audit your existing assets and decide what you actually need.
I started by looking at my Instagram analytics for the previous 90 days. My best-performing posts were lifestyle shots—someone holding a wallet, a cardholder on a café table next to a coffee cup, a tote bag on a wooden bench. My worst performers were the flat white-background shots. So I needed lifestyle images and short videos showing the product in use. I made a list: six wallet shots (different settings), six cardholder shots, six tote bag shots, and 12 videos (10–15 seconds each) showing how the cardholder slides into a pocket or how the tote bag looks when you're carrying it.
Don't skip this step. I wasted time generating a beautiful image of a wallet on a marble countertop before realizing I'd never posted marble-countertop content and my audience had no reason to expect it. Make what your audience already responds to, just better.
2. Use Stable Image Core for lifestyle product shots with simple backgrounds.
I used Stable Image Core on Kryotta for the first round of images. My prompt: "A brown leather bifold wallet on a wooden café table, morning light from the left, white coffee cup in soft focus in the background, product photography, shallow depth of field, Canon 5D Mark IV, 50mm lens." I generated four variations, picked the best one, and had a usable image in three minutes.
What worked: Clean backgrounds (wood, concrete, linen). Natural light descriptions. Specific camera references (the Canon 5D detail made the lighting more realistic). Products sitting on flat surfaces.
What failed: Anything with hands. I tried "a hand holding a brown leather wallet" and got six-fingered nightmares. I tried "a wallet resting on an open palm" and the palm had no wrist. Stability is excellent at objects and environments, but hands are still a disaster. If you need hands, hire someone on Fiverr for $15 and composite the product in later, or just avoid hands entirely.
3. Avoid text, logos and fine detail in AI-generated product shots.
I have a small logo embossed on the inside of my wallets—just my initials in a sans-serif font. I tried generating an image that showed the logo. Every single attempt gave me blurry nonsense or letters that looked like Cyrillic. I also tried generating a shot of a cardholder with three credit cards visible in the slots. The "cards" had random numbers and fake logos that looked like a legal liability.
Solution: Generate the background and setting, then composite your real product photo on top in Canva or Photoshop. Or just don't show text. My best-performing image was a tote bag on a park bench with fallen leaves around it—no logos, no cards, no text. It looked real because there was nothing for the AI to get wrong.
4. Use Luma Ray 2 for short product videos with simple motion.
I tested both Luma Ray 2 and Nova Reel for video ads. Luma was better for the kind of motion I needed: a slow pan across a wallet on a table, a zoom-in on a cardholder, a gentle tilt-up from a tote bag to show the full shape. My prompt: "A slow dolly-in shot of a brown leather wallet on a wooden table, morning light, product video, 4K, shallow depth of field." I got 10 seconds of usable footage in about 90 seconds of generation time.
What worked: Single-direction camera moves (dolly in, pan left, tilt up). Static products. Soft lighting. I used these videos as the first three seconds of Instagram Reels, then cut to real footage of me opening the wallet or showing the stitching.
What failed: Any video where the product moved. I tried "a hand picking up a wallet from a table" and the hand phased through the wallet like a ghost. I tried "a cardholder sliding into a jacket pocket" and the pocket opening looked like a portal to another dimension. Luma is great for camera movement around a static object. It's terrible at object movement or interaction.
5. Use Nova Reel for faster, simpler motion when you need volume.
Nova Reel is faster than Luma—about 40 seconds per video—but the motion is less cinematic. I used it for the second batch of videos where I just needed something moving on screen: a wallet sitting on a table with the camera slowly rotating around it, a tote bag with the "camera" circling 180 degrees. The quality was a step down from Luma (more artifacting, less realistic lighting), but when I tested them as TikTok ads, the engagement difference was negligible. People scrolling TikTok don't pause to admire your lighting.
I made six videos with Nova Reel and six with Luma Ray 2. The Nova Reel videos cost me less generation time, and two of them ended up in my top five performing ads by click-through rate. If you're making a dozen videos and you need them today, Nova Reel is the better choice.
6. Test every asset with a $5 Meta ad before you commit to the creative.
This is the step that saved me. I generated 18 product shots and 12 videos, but I didn't load them all into my Black Friday email or my Instagram grid. I ran a $5 Meta ad for each one—boosted post, one day, audience: people who'd visited my site in the last 90 days. I tracked click-through rate. The top eight images and top four videos got used in the actual campaign. The rest I deleted.
The worst-performing image: a wallet on a marble countertop (too fancy, didn't match my brand). The worst-performing video: a tote bag "swinging" as if someone were walking (the motion looked like the bag was haunted). The best-performing image: a cardholder on a wooden table next to a notebook and pen, warm afternoon light. The best-performing video: a slow dolly-in on a wallet, five seconds, no hands, no movement except the camera.
What actually converted during Black Friday week
I sent three emails during Black Friday week. The first one (Wednesday before Thanksgiving) used the wooden-table cardholder image. Click-through rate: 8.2%, up from my usual 4–5%. The second email (Black Friday morning) used a Luma Ray video of a wallet with a slow zoom-in, embedded as a GIF. Click-through rate: 9.1%. The third email (Cyber Monday) used a tote bag image with fallen leaves. Click-through rate: 6.8%.
On Instagram I posted four Reels during the week. Two used Luma Ray 2 video for the first three seconds, then cut to real footage. One used a Nova Reel video for the entire 12 seconds (the circling tote bag). One was entirely real footage. The two hybrid Reels (AI intro, real footage after) got 40% more shares than the all-real Reel. The all-Nova Reel got the fewest shares but the highest save rate—people wanted to come back to it, maybe because it looked different from everything else in their feed.
Total sales that week: $11,400, up from $6,200 the previous Black Friday. I can't say the AI assets caused the entire increase (I also ran better Meta ads and had more inventory), but the click-through rates on the emails and the engagement on the Reels were both measurably higher than my 2023 campaign.
What I'd do differently next time
I'd skip hands entirely. Every attempt at generating hands wasted time, and the images without hands performed just as well. I'd also generate twice as many assets and test them all with $5 Meta ads before committing to the hero image in my email. The image I thought would perform best (the marble countertop) bombed, and the image I almost deleted (the notebook and pen) was the winner.
I'd use Luma Ray 2 for any video that needs to look premium—email hero videos, Instagram Reels that I'm going to pin to my profile. I'd use Nova Reel for TikTok ads and Instagram Stories where speed matters more than cinematic quality. And I'd composite real product photos on top of AI-generated backgrounds whenever I needed to show text, logos or fine detail. Stability AI is incredible at making a beautiful wooden table or a sunlit park bench. It's not good at making legible text.
One more thing: I'd start two weeks earlier. I had six days, and I made it work, but I was testing and regenerating until 11 p.m. the night before I sent the first email. If you're reading this in October and planning for Black Friday, you have time to generate 50 assets, test them all, and pick the best ten. Don't wait until the week before.
Questions people ask
Can I use AI-generated product images on Etsy or Amazon?
Etsy's policy allows AI-generated images as long as you disclose that you used AI in the listing (check their current rules—they update frequently). Amazon requires that your main image show the actual product on a white background, but you can use AI-generated lifestyle images in the additional image slots. I'd still test them with a small ad budget first to make sure they don't look fake to your audience.
Do I need to tell customers the images are AI-generated?
Legally, it depends on your platform and jurisdiction (Etsy requires disclosure, Shopify doesn't). Ethically, I think it's fine not to disclose if the image accurately represents the product and you're not misleading anyone. If you're generating an image of a wallet on a table and you sell that exact wallet, it's no different from hiring a photographer to shoot the same scene. If you're generating a fake product or a feature your product doesn't have, that's a problem.
Which tool is better for video, Luma Ray 2 or Nova Reel?
Luma Ray 2 if you want cinematic quality and you're willing to wait 90 seconds per video. Nova Reel if you need speed and you're making ads for TikTok or Stories where people scroll fast. I used both, and both had videos in my top five performers. Test a few with each and see what your audience responds to.
What if the AI-generated image doesn't look exactly like my product?
Generate the background and setting with Stability AI, then composite your real product photo on top. I do this in Canva—drop the AI background, remove the background from my real product photo (Canva's background remover works fine), place the product on the AI background, adjust the lighting and shadows. Takes five minutes and looks more real than a fully AI-generated image.
I've been using Kryotta to run all of these models in one place—Stable Image Core for the product shots, Luma Ray 2 and Nova Reel for the videos, and I can compare outputs side-by-side before I commit to one. If you're planning a Black Friday or Cyber Monday campaign and you need to make a lot of assets fast, it's worth testing this process now while you still have time to fix what doesn't work. Start a free trial at Kryotta and generate your first batch this week.
