AI product images – are you using them? How’s it working out?

We’re experimenting with Flux and Stable Diffusion for more control. Midjourney is great but sometimes too “stylized.” If you want very specific angles, lighting, or product placement, Stable Diffusion with a LoRA trained on your product photos gives you way more consistency. Higher learning curve though – you need someone on the team who knows what they’re doing.

One thing nobody mentioned: AI is really good for generating multiple scene variations quickly. Instead of doing one lifestyle shot and hoping it works, you can generate 10 different versions, put them in a test, and see which one converts best. Then just have your designer polish the winner.
That’s been the biggest win for us – not replacing designers, but giving them more data to work with before they spend hours on something that might not resonate.

don’t rely on pure AI*
This. We tested this on a few products last year. AI-only images got the listings up fast, but conversion was noticeably lower than products with real photography. Customers can tell. If you’re selling in a category where trust matters (apparel, skincare, any premium positioning), hybrid is the way.

That said, for quick tests or high-volume categories where customers aren’t that picky, AI is a huge time-saver.

We’re a Temu semi-managed seller, similar team size (around 7 in ops). I feel your pain – we’ve had listings get pulled because we couldn’t get images back from design fast enough.
Here’s what’s working for us after a few months of trial and error:

1. Our setup:** Ops uses AI (Midjourney + a bit of Adobe Firefly) to generate base images – product on white, a few simple lifestyle scenes. Takes about 2 hours per set. Then one dedicated designer does final polish: color correction, shadows, templates. The designer’s role shifted from creating everything from scratch to being a “finisher.” Efficiency is up maybe 60%, and turnaround is usually 1–2 days now.
2. Can ops handle images on their own?** Yes, for some products. For test runs, clearance items, or anything we’re just throwing at the wall to see what sticks? Absolutely. AI images are good enough to get the listing up and start collecting data. You can always re-shoot later if the product takes off.

But for your main sellers – the products that actually make money – don’t rely on pure AI. The images often look “off.” Plastic-y lighting, weird shadows, inconsistent angles. Customers notice. When the product doesn’t match the photo, returns spike, and then your seller rating drops, and then your traffic disappears.

1. Would we pay for this?** Yes. We did the math: a decent full-time designer in a mid-tier US city runs $4,000–5,000/month with benefits. If a tool can automate 80% of the grunt work and let our designer focus on the 20% that actually drives conversions, we’d happily pay $200–300/month for the team. Above that, margins get tight.

My advice: Don’t expect a magic button. The best setup is ops generating “80% ready” images, then a designer handling the final polish. And for your top sellers, real photography is still the way to go.

AI works better for some product categories than others. Tech products, futuristic styles, abstract or cartoon designs? AI handles those really well. But beauty products, skincare, anything where texture and material matter? The AI-generated images look fake. You’ll get higher returns when customers receive something that doesn’t match the photo.
We use a hybrid approach: AI for initial concepts and backgrounds, then real photography for the hero shot, especially for products where material quality is the selling point.