White-bg master
Focus-stacked packshot, marketplace-compliant fill, geometry locked.
Jewelry is where generic image AI dies publicly: mirror-bright metal is exactly the surface generative models confabulate — hallucinated hallmarks, prong counts that change between generations, engravings that spell nothing, and white gold drifting into silver. 'Jewelry product photography' pulls ~170 US searches a month at difficulty 15, but the category's real economics are returns: a ring that photos larger than it ships is the #1 dispute driver, and a gemstone that renders more brilliant than the grading is a refund plus a chargeback. The Berrys graph starts from a real focus-stacked packshot and derives every marketplace slot, size-card, wrist/neck context shot and 360° spin from locked geometry — sparkle is composited from real refraction, never imagined.
Reviewed October 2026 · Prepared by Berrys · Technical specifications & Canvas node workflows
| Parameter / Slot | Marketplace & Production Requirement |
|---|---|
| Marketplace cover | Amazon: pure white RGB 255, piece ≥85% of frame, single unit, no model, no props · Shopee: solid bg, ≥60% fill, no text on cover, 1:1 |
| Scale defense | Slot 2–3 must carry a mm size card or hand reference; ring sizes in mm + US/EU conversion, chain lengths in inches/cm — generated from your real spec sheet, never the model's 'sense' of size |
| Metal color truth | 5000–5600K lighting profile, no warm filters; white gold vs sterling vs platinum proofed against physical swatches — ΔE drift is a return, not a style choice |
| Sparkle & facets | Highlight structure comes from real refraction (focus-stacked capture or physics-based render of the actual cut) — models smear facets into generic glitter |
| Macro depth | Focus stacking is the craft: a single-exposure macro render always shows the wrong plane sharp across prongs, clasp and engraving |
| Hallmarks & engravings | 14K/925 stamps and custom engraving are cropped from the real capture; anything that redraws them invents a metal grade the item doesn't have |
| Video | 360° turntable spin + macro sweep; keep the stone's movement physically tied to the real faceting |
Focus-stacked packshot, marketplace-compliant fill, geometry locked.
Real mm ruler + ring-size/chain-length table generated from your spec sheet.
Stacked close-up of stone/setting — the zoom layer is the trust layer at $80+ AOV.
Show what ships: lobster clasp, butterfly backs, engraving side. Cuts 'not as pictured' disputes.
Neck/wrist branch from the locked piece; skin and drape are secondary, the jewelry pixels are never regenerated.
Real box/pouch + certification panel cropped from genuine artwork — the only honest slot for grades.
Reflective metal means every generation sees different room-lighting in the bracelet — and models happily invent text in those reflections: hallmarks, logos, engravings that don't exist. An invented '925' on a plated item isn't an aesthetic bug, it's a misrepresentation. Reflection content is locked to the capture or painted out.
Prong count, link shape, stone count and bezel thickness change per prompt run — the classic AI-jewelry tell buyers have learned to spot. SKU lock keeps one physical truth across the whole set.
A ring on pure white reads any size the buyer wants; the moment it arrives, the review writes itself. Size-card and in-hand slots are mandatory workflow outputs, not optional extras.
One stacked macro session per design family; the graph derives every marketplace crop, context shot and spin from it.
Lighting recipe, metal profile and scene grammar pinned per collection so a drop of 40 SKUs reads as one line.
360° and macro-sweep clips from the locked reference — the movement you'd otherwise book a rig and a day for.
Already working in an AI agent? Connect Berrys to Claude, Cursor, ChatGPT, Codex, or Hermes. Your agent plans the jewelry & accessories suite and runs the shots on your Berrys canvas, with saved Brand Memory and credit approval in chat.
https://app.berrys.io/api/mcpRemote MCPUsing Berrys, create a 6-slot luxury jewelry PDP suite for my piece in ./assets: 1. File node: Ingest focus-stacked macro photo of jewelry. 2. Brand System node: Set neutral 5500K diffusion lighting. Lock metal profile (gold/silver/platinum) and gem facet count to prevent hallucinated stamps or prongs. 3. Image node (Nano Banana 2.1): Render 2048px hero packshot on pure white with sharp specular highlights. 4. Macro pass: Render 2048px extreme close-up on gemstone facets and setting prongs. 5. Scale card: Generate mm millimeter ruler comparison shot to prevent size-related returns. Estimate credit cost and request approval before generating.
Jewelry is a trust category where the image is a representation of graded, measurable goods. This workflow scales the presentation layer — crops, context, scale cards, spins, collection consistency — from real capture. It must never touch the claims layer: carat, clarity, metal stamps and certificate panels come from the physical product and its paperwork. A render cannot grade a stone; keep the grading in words and paperwork where it belongs.
It can, which is exactly why you shouldn't. Inflating apparent stone size or brilliance beyond the grading is a listing-truth problem; the graph keeps facet behavior tied to the real capture and lets the size card do the honest scale work.
Where the rule is 'accurately represent the product' (Amazon, Shopee), derived renders from real capture pass; invented ones that change prongs, stamps or stones will generate disputes and suppression.
At macro distances a single frame can't keep prong, girdle and engraving sharp together. Stacking is the craft standard; the canvas automates the derivative slots, not the truth capture.
Yes — the size-card node generates from your spec sheet (mm, US, EU), so the chart is data, not typography the model hallucinated.
Start free on Berrys Canvas — 50 credits, no card required — or send Berrys Studio a brief and get finished campaign visuals with agency art direction.