1 · In
Your catalog + swatches
Products, fabrics, finishes and source photos from Shopify.
Case study · Home & decor
AI product images for every fabric and finish
I built a production pipeline, over about six months, that applies fabrics and finishes across furniture to produce product images. Here's how it works.
Problem
Every fabric and finish needs its own image
Show one sofa in 20 fabrics and you need 20 sets of images. Multiply that across the catalog and you can't shoot every combination, so new options wait on a photo shoot.
Approach
A pipeline, not a one-off render
Products and source photos come in through the Shopify Admin API. The pipeline renders each fabric, finish and color across products, then writes alt text, descriptions and SEO metafields. Everything lands in a review queue, and nothing goes live until your team approves it.
Model comparison
Choosing an image model for product variants
Three questions decide the model: can it take the swatch as an image, can you mask just the area to change, and what does each image really cost? Here's what vendors published as of October 9, 2026.
| Model | Swatch as image | Mask | Price per 1K image |
|---|---|---|---|
| Gemini Nano Banana 2.1 | Yes, up to 14 refs | No (prompt only) | $0.0336 (batch $0.0168) |
| Gemini Nano Banana Pro | Yes, including 3 style refs | No | $0.134 (batch $0.067) |
| FLUX 3 Image Edit (fal) | Yes, 1–10 refs | No | $0.048 |
| FLUX.1 Kontext [pro] (fal) | No, text only | No | $0.04 |
| OpenAI gpt-image-2.5 | Yes, multiple inputs | Yes, but not pixel-exact | Token-priced, so measure it |
Notes. Every Gemini output carries a SynthID watermark. FLUX.1 Kontext [dev] is under a non-commercial license, and commercial terms for FLUX 3 Edit are still to be confirmed.
Quality: untested. Output quality, fabric fidelity, and consistency haven't been measured yet. I'm building a public demo repo that runs 3 products × 3 fabrics through each model and scores shape, texture, seams, cost, and speed. This section gets updated when it does.
List prices from vendor pricing pages, checked 2026-10-09.
Pipeline flow
From catalog to Shopify in three steps
2 · Generate
Variants + copy + SEO
Fabric and finish renders, image variants, descriptions and SEO fields.
3 · Publish
Review → Shopify
Your team approves every item before it goes live.
Results
Results: not published yet
Quality: untested. A public demo repo will run 3 products × 3 fabrics through each model and score shape, texture, seams, cost, and speed. No test grid or quality scores are published yet.
The demo-repo test grid goes here once the run is public.
Next step
Want this for your catalog?
The AI Product Content Pipeline is this build, scoped to your store: fixed price, five weeks, and your team approves every image.