Research & posts
What we learn building and selling jewelry online — measured, written down, published.
Library
/
August 2026
- Generate or Replace? What 240 Jewelry AI Images Taught Us — Two AI workflows, 240 images, one durable lesson: generating a new image and replacing a product in an existing image are different jobs — and they have different winners.
- Replace First, Compose Second — Our production method for AI jewelry imagery: lock the product's truth first, then build the scene around it. Six steps, one approval logic.
- The 8 mm Septum Ring Method — A case anatomy: keeping an 8 mm gold clicker accurate through a full AI image production chain — placement guide, exact replacement, approvals, and the shots that made it to the shop.
- We Ran a Census of 15,823 Jewelry Brands on Instagram. Followers Predict Almost Nothing. — Half the industry's Instagram accounts are effectively dead, small accounts convert followers to shop traffic 4–8x better than big ones, and engagement rate does not predict commercial reach.
- The Jewelry Shop Benchmark: What We Score on Product Pages, Shipping, Returns and Trust — The rubric we use to audit jewelry shops — the specific things that decide whether a visitor who likes the piece actually orders it.
- The Five Kinds of Jewelry Photography — Packshots, styled, on-model, editorial, authentic — the five shot types every jewelry shop needs, what each one is for, and why most shops only use one.
- Can AI Sell Jewelry? What We Check Before Trusting Assistants With Customers — AI assistants are starting to answer buyers' questions about sizing, materials and care. Here is where they help, where they hallucinate, and the test we run before trusting one.
- The Feed Tile Is the First Ad — A field note from running our own Google Shopping feed: the product tile does the selling before your shop ever gets the click.
- Creative Is Targeting: The Jewelry Version — Modern ad platforms find the audience your creative describes. In jewelry, the shot type is the audience selector.
- The Route From a Turkish Workshop to a European Customer — Turkey manufactures world-class jewelry that mostly sells as anonymous wholesale. The route to selling it directly to European customers — decisions, sequence, and honest hard parts.
- One Channel First — Why new jewelry shops should prove one acquisition channel before touching a second — and how to pick it.
- Choosing a Launch Market With Instagram, Search and Local Business Data — A scorecard method for picking your first country: measure demand, competition and feasibility with public data before spending a euro on ads.
- The Complete Jewelry Categories Reference — A comprehensive mapping of all jewelry types to body locations, organized by wear method. Foundational research for AI jewelry photography.
- Studio Product Photography: Evaluating AI Model Performance Across Camera Angles — Testing 4 frontier models across 5 camera angles reveals Generate workflow dominates for studio shots, with Nano Banana Pro achieving 93% publishable rate and perfect ring accuracy.
- From 0% to 100%: How We Achieved Reliable AI Camera Angle Control for Jewelry Photography — A practical case study showing how synonym stacking and hierarchical prompts took our AI-generated jewelry photography from 0% accuracy to 100% across five camera angles. Includes working prompts.
- Evaluating Frontier Image Generation Models for Commercial Jewelry Photography — A systematic comparison of 6 frontier models across 270 pairwise evaluations reveals workflow-dependent performance patterns with significant implications for production deployment.
- AI Jewelry Photography: Which Models Can Even Do It? (Part 1) — We tested 11 AI image generators to find which ones can produce usable jewelry photography. This elimination study identifies which models to avoid entirely and which deserve further testing.
- The Jewelry Photography Taxonomy — A foundational framework classifying all jewelry photography shot types.
- Prompt Composition — How template parameters, brand skin, and seasonal skin combine into AI prompts.
Work with StudioFormel
