Bottom line: a 12-person US home-goods DTC brand cut product photography cost by 85% with AI. The money was saved, but conversion dropped 11% three months later - and that part is more useful than the savings. Many small e-commerce shops see only the first half and crash on the second.
Background
The brand sells Nordic-style home accessories, 200-plus SKUs, new drops every quarter. Team of 12: 3 operations, 2 design, 1 part-time photographer, the rest on supply chain and support. Each SKU used a photo studio: set, lighting, retouch, about 400 dollars. Across 200 SKUs that is 80,000 dollars a year in photography alone, and cash flow got tight during launch seasons, often advancing money to wait for payback.
The founder is not technical, but was talked into AI image generation by peers, saw competitors "produce dozens of images a day," and greenlit it in early 2026 - switching fully without a small pilot test.
How they used AI
Step 1: Midjourney for concept scenes
With the brand palette and "Nordic living room" prompts, they generated usage-scene images in batches. One studio shot cost 400 dollars; Midjourney produced 50 scenes for a few dollars of amortized subscription. This drove scene-image cost near zero, and drop cadence moved from quarterly to monthly.
Step 2: Leonardo AI for white-background product shots
They fed a few real product photos into Leonardo's reference feature to generate uniform white-background, shallow-depth product images, replacing studio white shots. Batch, fast, style-adjustable - hundreds in an afternoon.
Step 3: Assemble the listing
Scene plus white-background images formed the listing; copy was written by AI. The whole flow went from "two weeks waiting for photography" to "images in two days." They briefly thought they had found free lunch for content production.
The pitfalls (the important part)
- Style drift: the same Midjourney prompt run ten times gave different leg angles, light, and materials, so the listing looked like different brands. Returning customers asked on social media, "did you switch suppliers?" - brand recognition diluted.
- Copyright and authenticity doubts: users questioned "is this a real product," because AI wood grain occasionally showed nonexistent seams and misaligned texture, hurting trust; one doubt post drew dozens of echoes.
- Conversion drop: A/B tests showed pure-AI listings converted 11% lower than real studio shots. Users said they "couldn't feel the texture" and "it did not match the item." High-ticket SKUs (above 200 dollars) fell hardest; low-ticket ones barely moved.
The most expensive part was not the saved money, but the three months of rework after the pitfalls. They once pushed 23% of SKUs back to redo, team morale suffered, and design shifted from "producing images" to "putting out fires."
How they fixed it
They did not return to full studio shooting, but built a hybrid flow: AI for scenes and drafts, but key SKUs still got one real studio "trust-anchor" shot as the first image, with AI extending the rest; high-ticket SKUs kept all-human images, low-ticket went full AI. Rework dropped from 23% to 7%, conversion returned close to studio levels, and cost stayed 60% below full studio.
What this means for you
Small and mid-size e-commerce can absolutely use AI to cut content cost, but do not replace everything. First-image trust, texture, and brand consistency are still human work for now. Use AI as an amplifier, not a replacement. One actionable play: use AI to test volumes of variants first, then add human retouch only to SKUs that prove out in the data.
ROI recap
Per-SKU photography fell from 400 to 60 dollars, saving 38,000 dollars in six months; the cost was three months of conversion experiments and rework, plus a small test of brand trust. Net it still paid off, but the "all-AI" route did not hold. The pragmatic play: AI for volume, humans for quality. Once that line is clear, AI is a tool that earns you money, not a shortcut that digs you a hole.