I made 35 product shots for WhatsApp commerce with Stability AI then real photos: which got more orders

The WhatsApp status that sold three handbags in twenty minutes

I posted a photo of a leather crossbody bag to my WhatsApp status at 11 a.m. on a Tuesday. The image was terrible—I'd taken it in my Harare flat with the curtains half-closed, the bag propped against a sofa cushion that had a visible coffee stain in the corner. The leather looked dull. The stitching barely showed. By 1 p.m. I'd sold zero bags and fielded two messages asking if the colour was "really that flat" or if my camera was just bad.

The next day I generated a product shot with Stability AI. Clean white background, the bag floating at a slight angle, stitching crisp, leather grain visible. I posted it at the same time, 11 a.m. By 11:20 I had three EcoCash payments confirmed and five people asking if I had it in brown.

That result made me wonder: if AI images work that well, why bother with real photos at all? So I ran a proper test. I sell imported accessories and homeware via WhatsApp—bags, wallets, picture frames, small kitchen gadgets—mostly to customers in Harare and Bulawayo who find me through status updates or referrals. Over thirty days I posted 35 products twice: once with an AI-generated image, once with a phone photo. I tracked which version got more orders, how many people asked "Is this real?", and whether the data cost of generating images ate into the profit. Here's what happened.

How I set up the test without wrecking my sales

I didn't want to confuse my existing customer base, so I split the test across two WhatsApp Business accounts. Account A posted AI images; Account B posted phone photos. I invited half my contact list to follow Account A's status and half to follow Account B. Both accounts listed the same products at the same prices in US dollars, accepted EcoCash and cash on delivery, and responded to messages within an hour.

For the AI images I used Stable Image Core in Kryotta. My prompt template was: "Product photo of [item description], clean white background, soft studio lighting, sharp focus, professional e-commerce style, no text or watermarks." I kept it simple because I was generating these on my phone during load-shedding, tethered to mobile data. Each image took about fifteen seconds to generate and cost roughly the same data as loading two Instagram photos—not free, but cheaper than driving to a friend's place with better lighting to shoot real photos.

For the phone photos I used my Samsung Galaxy A14. Natural light from a window, plain wall or white bedsheet as background, no filters. I'm not a photographer, but I'm not hopeless either. The photos were honest: you could see the product clearly, but you could also see that I'd taken them in a normal room.

I posted each product once per account over the thirty days, then waited to see which image pulled more orders.

The results: AI won on speed, real photos won on trust for certain items

Here's the breakdown by product type:

AI images sold more handbags and wallets by a clear margin. The clean backgrounds made the stitching and hardware stand out. People messaged asking about size and colour, not whether the bag was real. The eight "Is this real?" questions came from new customers who hadn't bought from me before; I sent them a quick video of the actual bag and they all converted.

Phone photos sold more picture frames and jewellery. For frames, people wanted to see the texture of the wood or the exact shade of the finish, and the AI versions looked too perfect—several people said they "looked like catalogue photos, not what I'll actually get." For jewellery, the AI struggled with reflections and detail. A pair of silver earrings came out looking like plastic in the generated image. The phone photo, even with worse lighting, showed the metal properly.

Kitchenware was nearly even, but the AI images got twelve "Is this real?" questions versus one for phone photos. People wanted proof I had the item in stock, especially for higher-value things like a $28 blender. I started including a second image—a quick phone snap of the product in my hand or on my table—alongside the AI shot. That dropped the questions to three and kept the conversion rate high.

What the "Is this real?" question actually costs you

When someone messages "Do you actually have this?", you've already lost momentum. In a WhatsApp commerce model where buyers can't touch the product before paying EcoCash, that question is them deciding whether to trust you. If you don't answer fast, they move on. If your answer sounds defensive, they move on. If you take too long to send proof, they've already messaged two other sellers.

I tracked response time for those questions. On average, it took me four minutes to reply with a verification photo or video. In that four minutes, I estimate I lost about 30% of potential buyers—they just never replied after I sent proof. The phone photos triggered far fewer trust questions up front, which meant fewer chances to lose the sale in that gap.

For new customers especially, the AI images required more reassurance. Repeat buyers didn't care; they knew I'd deliver. But someone finding me for the first time through a referral needed to see something that felt real. A slightly imperfect phone photo did that better than a flawless AI render.

When AI images are worth the data cost (and when they're not)

Generating 35 AI product shots over the month cost me about $2.50 in mobile data, assuming I was tethered and not on Wi-Fi. That's negligible. The real cost was the extra time spent reassuring buyers and sending follow-up photos for high-value or detail-sensitive items.

Use AI images when:

• You're selling bags, shoes, wallets, or clothing where shape and colour matter more than texture.
• Your phone photos are genuinely bad—poor lighting, cluttered background, shaky focus.
• You want to test a product's appeal before you've imported stock. I generated images for three handbag styles I was considering ordering from a supplier in Johannesburg. The AI shots let me gauge interest via status views and messages before I committed $300 to inventory.
• You're posting the same item repeatedly and want visual variety. I generated three angles of one crossbody bag and rotated them across different status updates. It looked like I had better photography skills than I do.

Stick with phone photos when:

• You're selling jewellery, electronics, or anything where material and finish are the main selling points.
• Your customers are new and trust is fragile. A real photo, even if it's not perfect, signals "I have this item in my hands right now."
• The product has visible branding, logos, or text that AI might render incorrectly. Stability AI occasionally invented letters on a watch face I tried to generate. The phone photo showed the real dial.
• You're doing cash on delivery and people want proof you're not a scam account. Real photos with your environment visible—a corner of your table, a bit of your floor—make you human.

The hybrid approach that worked best by week three

By day twenty I'd stopped treating this as AI versus real photos and started using both. My new workflow: generate a clean AI product shot for the main status post, then keep a phone photo ready to send in chat when someone asks for details or proof of stock. The AI image gets attention and makes the product look appealing. The phone photo closes the deal by proving I'm not dropshipping from China or running a scam.

For high-ticket items—anything over $40—I now post the AI image with a caption that says "DM for actual photos" or "Video available on request." That filters out time-wasters and signals to serious buyers that I'm transparent. Conversion rate on those items went up because the people who messaged were already interested, and the follow-up photo removed their last objection.

I also started using Stable Image Core to fix bad lighting in real photos rather than generating from scratch. I'd take a phone photo, then use the image-to-image mode in Kryotta with a prompt like "Improve lighting and remove background, keep product exactly as shown." That kept the authenticity of the real item while cleaning up the presentation. It worked well for kitchen gadgets and frames where texture mattered but my flat's lighting didn't cooperate.

What I'd do differently if I ran this test again

I'd track not just orders but also repeat customers. I didn't separate first-time buyers from people who'd bought from me before, and I suspect the trust gap for AI images is much bigger with new customers. If I ran this again I'd split the data that way.

I'd also test video. WhatsApp status supports short videos, and a five-second clip of me holding a handbag or turning a picture frame would probably outperform both AI and static phone photos. I didn't try it this time because I assumed video would take longer to produce, but honestly it might be faster than setting up a decent photo.

And I'd generate AI images in batches when I had Wi-Fi, not on mobile data during the day. The data cost was small but it added up, and I wasted time waiting for images to load when I could have prepped everything the night before.

Questions people ask

Can I use AI product images if I'm selling second-hand or used items?
I wouldn't. The whole point of a used item is showing its actual condition—scratches, wear, patina. An AI image will either hide that (making you look dishonest when the buyer receives it) or fail to render it accurately. Stick with real photos and be very clear about flaws.

Do AI images work for food or perishable goods?
I didn't test this, but I'd be cautious. People want to see the actual food you're selling, especially if you're doing home baking or prepared meals. An AI-generated cake might look perfect, but if your real cake doesn't match it you'll get complaints and refund requests.

How do I stop people thinking I'm dropshipping when I use AI images?
Include one real photo in your status or chat. A quick snap of the product on your table, in your hand, or with a piece of paper showing today's date. It takes ten seconds and kills the dropshipping suspicion immediately.

Which Stability AI model should I use for product photos?
I used Stable Image Core because it's fast and the output is clean. I tried SD 3.5 Large for a few images and the detail was sharper, but it took longer to generate and I didn't notice a difference in sales. For WhatsApp commerce where people are viewing on phone screens, Core is enough.

If you're selling on WhatsApp and your phone photos aren't converting, try generating a few AI product shots in Kryotta and see if your order rate changes. Start with items where appearance matters more than texture, keep a real photo ready for follow-up, and don't be afraid to mix both approaches. You can test Stable Image Core and the other models at app.kryotta.ai/auth—it's faster than driving across Harare to borrow someone's ring light.