AI Image Analysis — Auto-Generate Properties & Descriptions from Product Images

About the Extension

Turn product images into structured data, copy and SEO meta tags

Turn product images into structured data, ready-to-publish copy, and SEO meta tags in seconds.

This plugin uses Anthropic Claude or OpenAI GPT-4o Vision APIs to automatically extract every relevant attribute from a product's cover image: colour, material, style, designer, manufacturer, era, technique, origin, full description, SEO meta title — and whatever else you need.

Particularly valuable for image-heavy catalogues: vintage, antique, design, fashion, jewelry, art, classic cars — anywhere the picture is the primary source of information and manual curation eats hours per product.

What you get

• Time saved: 1000 products in hours, not weeks
• Clean filters: Normalised colour palette (21 colours including metals like gold, silver, chrome) — no more "forest green" vs "pine green" duplicates
• Designer AND manufacturer separately — e.g. "Arne Jacobsen" as designer and "Fritz Hansen" as manufacturer, both as independent property groups
• SEO automation (optional) — the AI generates a matching meta title (~60 chars) and meta description, written directly into Shopware's native SEO fields
• Product context in the prompt — product name and Shopware manufacturer are passed to the AI, so e.g. "Grundig" is detected from the product name even when no logo is visible
• Fully configurable: You define the fields you want extracted — perfect for your vertical
• Approval workflow: Every AI result starts as "pending" — review and approve, or fix inline with a few clicks
• Transparent costs: You pay the AI provider directly (~1 ct per product). No markup, no subscriptions on our end.

Technically

• Image auto-resized to 1024 px (cost optimisation)
• Properties created as Shopware Property Groups with translations
• Description stored in an HTML custom field on the product
• SEO meta title + description optionally written directly to the native Shopware fields
• Provider switch Claude ↔ OpenAI without code changes
• CLI, admin button, and REST API all included

No monthly fees, no hidden costs — one-time purchase. API costs go directly to Anthropic/OpenAI.

Features at a glance

• AI Vision — Claude 4 Sonnet/Opus or GPT-4o, switchable
• 12 default fields — colour, designer, manufacturer, material, style, pattern, technique, era, design reference, origin, meta title, description
• Define custom fields — brand, size, fit, year, gemstones, artist … any vertical
• Normalised colour palette with 21 colours including metals (gold, silver, chrome, brass, copper, bronze)
• Per-field confidence — certain / uncertain / manual
• Multi-value fields — e.g. Material: ["Metal", "Faux leather"]
• SEO automation — meta title and meta description optionally written directly to Shopware's native SEO fields
• Product context in prompt — name + SKU + Shopware manufacturer passed into the analysis
• Approval workflow — pending → approved / rejected
• Inline edit in product detail — fix individual fields with one click
• Batch analysis via CLI — with category filters and rate limiting
• Preview mode (dry-run) — test without saving
• REST API for custom integrations
• Variant support (falls back to parent cover)
• Bilingual (DE/EN), auto-translation across all shop languages
• Deduplication — property groups are reused, no duplicates
• Auto-resize to 1024 px for API cost optimisation

Perfect for

• Vintage & antique shops — designer, era, style straight from the image
• Furniture & interior — material, style, provenance
• Fashion — brand, material, fit, season, occasion
• Jewelry & watches — material, gemstones, style, era
• Art & galleries — artist, technique, motif, movement
• Classic cars & parts — make, model, year, condition
• Dropshipping & retail — any catalogue too large to curate manually

Why this plugin?

Shopware's native Copilot (Rise plan only) suggests properties from existing description text — but that requires well-maintained copy as a starting point. We take the opposite, more practical route: the image is the source. Upload product images, receive structured data + copy back.

Advantages over the built-in Copilot:

• Works without a Rise licence (usable on any Shopware plan)
• Generates from image, not text — no chicken-and-egg problem
• Specifically normalised colour palette with metals for shop filters
• Vertical-specific field configuration
• Approval workflow with inline edit
• Choice between Claude and GPT-4o