When a beauty brand came to us in late 2025 needing 200 product images across 40 SKUs for a catalog relaunch, they had two weeks and a budget that wouldn't have covered a single day in a traditional studio. We delivered 218 final images in 11 days using a hybrid workflow -- real reference shots, AI generation, and human compositing. That's the kind of thing that genuinely changes what's possible for mid-size e-commerce brands. But I want to be clear about what 'AI product photography' actually means in practice, because there's a lot of marketing fog around this topic.
What AI handles well: background work and clean catalog shots
Background removal has been solved for a few years now -- tools like Remove.bg and Photoshop's AI selection are essentially perfect for most products. What's newer is background replacement and scene generation. We use FLUX.1 for most of our background compositing work. You provide the isolated product, specify the environment (marble surface, soft shadow, warm ambient light), and the model generates a photorealistic background that the product sits in. Done well, it's indistinguishable from a studio shot.
Clean white-background product shots for marketplaces like Trendyol or Amazon are even simpler. We photograph the product once against a neutral background, remove the background, and use AI to generate the exact white/off-white gradient the platform requires. For catalog work, we can turn one reference photo into 8-10 variants -- different backgrounds, different lighting moods, different surface materials -- in an afternoon. That's not hypothetical. We processed 127 product variants in a single day last February for a home goods client.
Lifestyle shots without a studio: possible but product-dependent
This is where results vary a lot depending on what you're selling. For packaged goods -- bottled skincare, supplements, boxed products -- AI lifestyle generation works very well. The product shape is regular, the surface is mostly reflective but not hyper-detailed, and compositing the product into a generated lifestyle scene produces convincing results.
For fashion, it's more complicated. Apparel draped on a body requires either a real model, a 3D garment simulation (which is its own production pipeline), or accepting that AI-generated fashion looks slightly off when the fabric folds don't follow physics correctly. We've found AI works better for fashion accessories -- bags, shoes, jewelry on a surface -- than for clothing on a model.
Food photography is a mixed story. Hero shots of packaged food products work great. Close-ups of actual food -- a cross-section of a pastry, steam rising from a bowl, the texture of a cut of meat -- still look AI-generated to anyone paying attention. Real food photography is about physical texture and the behavior of real light, and AI is not consistently reliable there yet.
What still requires real photography
- Jewelry with gemstone detail: AI consistently misrepresents facets, reflections, and the specific way diamonds scatter light. Clients will notice.
- Textured handmade products: ceramics, leather goods with natural grain, handwoven textiles -- the surface texture is the product story and AI doesn't capture it accurately.
- Food macro and close-up shots: the difference between real food and AI food is visible at close range.
- Products where scale and proportion are critical: AI sometimes subtly distorts dimensions in ways that mislead customers about what they're buying.
- Any product requiring legally accurate representation for regulated categories (pharmaceutical, medical devices, safety equipment).
The actual cost comparison
| Scenario | Traditional Studio | AI-Assisted (Our Workflow) | Time |
|---|---|---|---|
| 20 SKUs, white background catalog | AI: 1-2 days vs. Studio: 1 day + editing week | ||
| 40 SKUs with lifestyle variants (3 per SKU) | AI: 5-7 days vs. Studio: 2-3 weeks | ||
| Single hero campaign image (1 final) | Similar turnaround | ||
| 100 product variants, e-commerce ready | AI: 8-10 days vs. Studio: 4-6 weeks |
Those AI-assisted prices include our retouching and compositing time. Pure generation costs (compute + software) are a fraction of that -- maybe 15-20% of the total. The majority of the cost is still human labor: art direction, curation, quality control, and the client feedback loop. Anyone who tells you AI production is 'nearly free' is either not doing quality work or not counting the labor honestly.
Tools we actually use
FLUX.1 Dev and Pro for most product compositing -- the photorealism and light handling are better than Midjourney for this specific use case. Midjourney v7 for lifestyle concept work and moodboarding. Adobe Firefly for in-workflow background generation when we're working in Photoshop anyway. Our own brand-trained LoRAs built on FLUX for returning clients where consistency across a catalog matters. We don't have a religious attachment to any single tool -- what we use depends on the product type and the brief.
A realistic workflow for 100 product variants
Here's how a recent 100-variant project actually ran. Day 1: client ships us 20 products, we photograph each against a controlled neutral background (2-3 angles each, 1-2 hours per product cluster). Day 2: background removal, product retouching, building the asset library. Day 3-4: LoRA training on the product set, generating background and lifestyle variants for approval. Day 5: client approval round. Day 6-7: revisions, final compositing, format delivery. That's the real timeline. Not 'AI does it in minutes' -- but also not a six-week studio production.
Frequently Asked Questions
Do you still need reference photos for AI product photography?
Yes. For any serious production work, we photograph the actual product first. The reference photos give us accurate color, proportion, and surface texture to base the AI generation on. Clients who ask us to 'generate from just the product description' will get generic results that don't accurately represent what they're selling.
Can AI product images be used on Amazon or major marketplaces?
Generally yes -- marketplace terms focus on image accuracy (the product must match what's shown) rather than how the image was produced. The practical issue is quality: marketplace product images go through automated quality checks and customer scrutiny. AI images that pass that bar are fine. Low-quality generations that misrepresent the product are not fine, for the same reasons a blurry or misleading studio photo isn't.
What about products with branding and text on the packaging?
This is still a challenge. AI models handle text on packaging imperfectly -- letters get blurred, rearranged, or distorted. Our standard approach is to generate the background and lighting environment with AI, then composite the real product (photographed accurately) into that environment. That way, the logo and packaging text are always accurate.
If you have a product catalog that needs photography and want to understand whether AI production makes sense for your specific products, send us a few examples. We'll give you an honest assessment -- including if traditional photography is the better choice.