How to photograph 700 products without a photoshoot.

The starting problem

A mid-sized promotional-goods distributor. 700 active SKUs on the shelf, around 200 of them seasonal. A Shopify store in preparation. No photo budget — or more precisely: the budget on the table is enough for a studio shoot of maybe 80 SKUs. The other 620 would get populated with manufacturer thumbnails that look inconsistent and don't match the brand.

This isn't an edge case. It's the default profile in B2B distribution: 500 to 2,500 SKUs, physical stock, no content production in-house.

Why a studio shoot doesn't scale

A product photographer charges between €15 and €40 per SKU. Floor: plain-white background, one angle. Ceiling: lifestyle scene with props and retouching. For 700 SKUs that's €10,500 to €28,000 — before you add scheduling, transporting the stock, and revision rounds.

The second wall is time. A photographer can shoot 40 to 60 SKUs per day. 700 products means 12 to 18 days of pure studio time. In that window the assortment changes.

The alternative path

One phone photo on the warehouse floor as a reference. An AI image model that turns that reference into a studio-grade packshot. A human review gate that checks every image before it ships. A connector that publishes to the shop.

Sounds like hype. It isn't — if you keep the four steps cleanly separated.

Step 1 — Intake on the warehouse floor

A warehouse worker walks the aisles with their phone. For each SKU: the SKU code (or a barcode scan), one photo from the top or front, and optionally a category note.

The quality of the phone photo matters less than you'd think. Crooked angle, bad lighting, plastic reflections — all tolerable as long as the product is fully visible and in focus. The AI model reads identity, proportions, and labelling from that photo. The aesthetic comes later.

Average intake speed: 80 to 120 SKUs per hour. 700 products takes 6 to 9 working hours, spread across 2 or 3 days. One warehouse worker, no photographer.

Step 2 — Process the reference image

The AI model gets two inputs: the phone photo and a descriptive prompt (category, positioning, lighting setup). The prompt isn't what most people assume. It doesn't describe the desired outcome in feeling words ("beautiful", "professional") — it describes the physical lighting situation: soft softbox, Phase One camera, seamless cyclorama cove, 100mm macro.

Why this works: the model has learned that when it sees those descriptors, it should render lighting physics. It doesn't invent "beauty"; it produces a plausible shot of a plausible setup.

Step 3 — The review gate

Here's the decision point that kills most AI image pipelines: who checks the output?

Naive pipelines push straight into WooCommerce or Shopify. Around 60% of images are usable on the first try. The other 40% — wrong proportions, invented labelling, unrealistic shadows — land unsorted in the shop. The shop manager only notices once a customer complains.

The fix: a review interface before the shop. Every SKU lands in a batch view. The operator sees the source photo, the generated version, and clicks Pass or Fail. On Fail: a short comment ("background too dark", "labelling wrong"), and the system regenerates with that comment injected as an extra rule.

Time per review click: 15 to 25 seconds on Pass, 30 to 45 seconds on Fail with comment. For 700 SKUs × 2 images that's roughly 8 to 14 hours of focused work — spread across 3 or 4 days.

Step 4 — Publish

Once an SKU is approved, the connector publishes: product title, description, category mapping, main and lifestyle image. Into collections on Shopify, into product categories on WooCommerce.

Time per SKU: about 4 to 8 seconds of REST latency. For 700 products: 45 to 90 minutes, one-off.

Real numbers from the example project

• Intake: 700 SKUs in 7 hours of warehouse work, spread across 2 days
• Generation: 1,400 images (packshot + scene per SKU) in 18 hours of compute
• Review: 11 hours of operator time, 29% fail rate in batch 1, 14% by batch 3
• Publish: 72 minutes for all approved SKUs
• Otto total: Pack L at €1,199 (up to 1,000 SKUs included)
• Studio comparison: €12,500 at €18 per SKU, not counting scheduling and logistics

What doesn't work

This pipeline isn't universal. Small metal parts with glossy surfaces, textiles with specific weave patterns, and food with strict colour-fidelity requirements are the three categories where we consistently see worse output. More detail in AI product photos — when they work, when they fail.

What this means

A distributor with 700 SKUs can have a populated, visually consistent shop in 10 working days. Price tag: €1,199. No photographer, no studio, no logistics. Two years ago this wasn't possible. Today it's routine work — as long as you don't rip out the review gate.