I own a women's clothing store on Wildberries and Ozon. By early March 2026, we needed to launch our spring collection — 200 SKUs. The budget for a studio shoot (photographer, model, retouching, locations) was around 500,000₽ and a three-week turnaround. That's a killer for inventory turnover: while you're shooting, competitors are already selling.
I tried AI as an experiment on 20 items. Two weeks later, the entire catalog was shot entirely with neural networks. Budget: 12,000₽ in Quantium credits. Quality? Studio-level. Customers on the marketplace can't tell the difference. Below, we'll show you how we did it, which models work for which shot types, and what you need to know about image rights.
The Problem: 200 SKUs and a 500k Studio Bill
Marketplaces need at least 4-5 shots per listing: a main shot on a white background, backstage/lifestyle, fabric detail, a full-body model shot, and an interior shot. For 200 SKUs, that's over 1000 images. A studio with 4 people shoots 20-30 SKUs a day on average. That means 10 days of work plus a week for retouching.
Cost in Moscow: 1500-2500₽ per item, including all shots and retouching. Let's take the average: 2000₽ × 200 SKUs = 400,000₽. Plus, shipping goods to the studio and back. That's a deal-breaker for a mid-sized business with a 30% margin.
Step 1. Listing Descriptions via ChatGPT
Before we tackled visuals, we ran all our descriptions through ChatGPT-5. Each item needed: a title up to 60 characters, a description up to 1000, key features, and answers to common questions. Our copywriter used to do 10 items a day; now it's 50. Here's the prompt:
Step 2. White Background Catalogue Shots via GPT-Image
The main shot is the product on a white background, following the 3:4 marketplace standard. We used GPT-Image with image-to-image for this: you upload a real photo of the item (even from your phone), then ask it to "place" it on a studio white background with proper lighting.
Here's the prompt formula we figured out after a week of testing:
GPT-Image holds the shape, color, and fabric texture 90-95% of the time. Logos and brand labels it draws poorly — so for branded items, we added a final shot with the real label using Photoshop, taking 3 minutes per item.
Step 3. Lifestyle Scenes via FLUX 2 Pro
A lifestyle shot means "the item on a model in a real setting." This used to be the priciest part of a shoot: model, location, weather, permits. Now? It's FLUX 2 Pro.
Here's the prompt template:
You just change the color and clothing type; the rest of the prompt keeps a consistent brand aesthetic. For 8 credits, you get a shot that used to cost 8,000-15,000₽. Learn more about working with FLUX in our 12 techniques guide.
Important: the model's face in AI photos shouldn't repeat from listing to listing (it makes the store feel "fake"). We created 3-4 different model looks and distributed them across our listings.
Step 4. Video Carousels via Sora 2
On Wildberries and Ozon, video listings boost conversion by 15-30%. Making videos for every SKU used to mean another production nightmare. We used Sora 2: 5-10 second clips showing the product in action.
Example for a dress:
For 15-25 credits, we get an ad-quality video that drives CTR on the marketplace. Not for all SKUs, but definitely for the top 30 bestsellers. For more on working with video models, check out our Sora vs. Veo comparison.
Step 5. Auto-Generate SEO Meta
On Ozon, Wildberries, and our own site, every listing needs a title, description, alt tags for photos, and an FAQ block. Our copywriter used to spend 20 minutes on a full SEO block for one SKU. Now, ChatGPT-5 does it in 30 seconds, based on the product description and a list of target keywords.
The prompt gives all the platform-specific fields at once: title up to 70 characters, description up to 160, 10 alt texts for images, 5 Q&A for the listing, and tags. That's 60 hours of copywriting saved for 200 SKUs.
What It Costs
| Shot Type | Studio (₽/SKU) | AI (₽/SKU) | Model |
|---|---|---|---|
| Main on white background | 800 | ~15 | GPT-Image |
| Lifestyle on model | 800 | ~20 | FLUX 2 Pro |
| Fabric detail | 200 | ~10 | FLUX 2 Pro |
| Interior shot | 200 | ~10 | FLUX 2 Pro |
| Video clip 10 sec | 3000 | ~80 | Sora 2 |
| Total per 1 SKU | ~2000 | ~50 | — |
For 200 SKUs, a studio costs 400,000₽. AI? Around 12,000₽ in Quantium credits. That's a 388,000₽ saving, or 33 times cheaper. Timeline-wise: 3 weeks for a studio vs. 5 days with just one operator.
A Quantium Premium subscription with 30,000 credits covers a medium-sized store's catalog update needs for 2-3 months.
The Legal Bit: Image Rights
This is a question you should ask right away. As of 2026:
- FLUX 2 Pro, GPT-Image, Gemini Image — commercial use is allowed with paid plans. With Quantium, rights to generated images belong to the user.
- Marketplaces (Wildberries, Ozon) don't formally prohibit AI photos, but they do require the image to match the actual product. If you use a lifestyle shot with an AI model, the item on her must match the real one.
- Advertising
- Don't try to do everything with AI. For your main marketplace shot on a white background, use a real photo or an AI image-to-image based on a real photo. No "conceptual" interpretations of the product.
- Lifestyle shots save the most money. AI shines here, delivering 30-50x savings without losing quality. If you're on a tight budget, start with lifestyle.
- Standardize your brand prompts. A single prompt template with a fixed color palette, lighting type, and style makes your whole catalog visually consistent. That's more important than making each shot "individually beautiful."
- Test with small batches. Before rolling out 200 SKUs, we ran 10 through AI, measured reactions, and adjusted prompts. That saved us two weeks and prevented a ton of revisions.
- Have a backup plan. If AI doesn't give you what you need for a specific item (this happens with 10-15% of SKUs due to complex geometry or texture), just reshoot it on your phone with good lighting. Don't force AI at all costs.
In six months of working with an AI stack, we boosted our profit margin by 8%. That's purely from saving on shoots and updating visuals more often. Our catalog now refreshes every two weeks, not just once a season, because it's technically possible. And that shifts our competitive edge on the marketplace more than you'd think.
Another less obvious benefit: testing hypotheses faster. Before, if I wanted to see if a new dress design would resonate with our target audience, I'd have to: sew a sample (3 weeks), shoot it (1 week), then run ads (2 weeks of testing). That's a month and a half until we got data. Now: AI generates a sample in an hour, AI photos are ready the next day, and ad tests start two days later. A month and a half turns into three days. At this speed, competitors using studio shoots just can't keep up.
Related resources: marketer case study, FLUX prompt guide, Midjourney alternatives.
Frequently Asked Questions
Can I use AI photos for product listings on Wildberries and Ozon?
Yes, marketplaces don't ban AI photos, but they do require the product in the image to match the real one. Here's the main rule: you can use AI photos for lifestyle and main product cards, but the required white-background product shot on each card must be accurate—ideally a real photo, or an AI photo that precisely replicates the actual product.
How much does AI save compared to studio shoots?
Studio shoots typically cost 1500-2500₽ per SKU (lighting, photographer, model, retouching). AI photos in Quantium, using FLUX 2 Pro and Gemini Image, run about 50₽ per SKU with a reasonable number of iterations. That's 30-40x savings on a batch of 200 SKUs—from 290,000 to 490,000 rubles.
Who owns the rights to AI-generated images?
FLUX, Gemini Image, and GPT-Image allow commercial use on paid plans. In Quantium, users own the rights to their generated images; you can freely use them in product listings, ads, and marketing. This is stated in the providers' terms.
What are the best models for catalog shots?
For white-background product shots, use GPT-Image (it's better at maintaining object identity from a reference). For lifestyle scenes with models or interiors, FLUX 2 Pro (for natural photorealism). For dynamic video carousels, Sora 2 (for 5-10 second clips of products in action).
How do I describe a product so AI reproduces it accurately?
Use image-to-image: upload a real product photo and ask the model to recreate it in your desired context. This works with GPT-Image and Gemini Image. Shape, color, and logo retention accuracy is 85-95%, meaning Photoshop retouching takes 5 minutes instead of an hour.
Try Quantium for free
20 credits per month on the free plan. 30+ AI models in one Telegram bot.
Open bot →Read also
