The brief: from a folder of artwork to a stocked Shopify store, automatically.

The client sells digital art and posters and needed to eliminate the most repetitive part of running the store: turning raw artwork files into complete, polished, SEO-ready Shopify product listings. The goal was a pipeline where simply dropping images into storage results — with no manual writing or data entry — in correctly categorised, well-described products appearing in the store. We built this as a 34-node n8n workflow split into two clean sub-workflows, using Airtable as the central database and status tracker so every image's journey from "unused" to "posted" is visible and controllable.

Sub-workflow 1 — Digital image analysis. The process begins with the artist uploading artwork to Google Drive and adding a row to an Airtable "raw image" table with the Drive file ID and a status of Unused. The workflow fetches all unused images, downloads each one from Drive, and passes it to an AI vision model that acts as an expert art analyst. Rather than a vague description, it extracts a precise, structured set of attributes: the character name, the series, the category (e.g. anime), any poster text, and the poster type or visual mood — all as clean JSON.

Crucially, the analysis is engineered against hallucination. The prompt explicitly instructs the model to focus only on the artwork content and to ignore any mockup context — frames, shadows, walls, or furniture — and to return null for anything unclear rather than inventing details. This discipline is what makes the downstream listing content accurate. The extracted attributes are written back to Airtable and the image is marked as processed, creating a continuous, self-advancing loop.

Sub-workflow 2 — Shopify product creation. With analysed data in hand, the second sub-workflow turns attributes into a sellable listing. First, it calls the Shopify Admin API to retrieve the store's existing collections, so each product can be matched to the most relevant category that actually exists in the catalogue — not an invented one. The analysed image data and the available collections are then handed to an LLM chain (running Gemini) that generates the full listing: a product title, an engaging and SEO-friendly description, the matched collection and its ID, and complete SEO metadata including page title, meta description, and URL handle.

A practical reliability detail sits here: the chain uses an auto-fixing output parser wrapping a structured schema, so if the model's first attempt produces malformed JSON, it's automatically corrected to fit the required structure rather than breaking the run. The generated listing content is saved back to Airtable, the product is created in Shopify via the native node, and the row's status is updated to indicate it has been posted. Throughout, batching, limit, and wait nodes pace the work to avoid hammering the APIs.

Why this design works. Splitting analysis from listing creation — each driven by Airtable status flags — makes the system both robust and observable: images can be analysed in one pass and turned into products in another, and the database always shows exactly where each item stands. Grounding the listing generation in the store's real collections means products land in the right place automatically, while the structured, anti-hallucination image analysis ensures the AI writes about what's genuinely in the artwork. The auto-fixing parser and throttling controls are the kind of pragmatic touches that keep a high-volume e-commerce pipeline running cleanly in production. The result is a true "upload and forget" system: digital artists, print-on-demand sellers, and store managers can drop images into Drive and watch fully structured, SEO-optimised products populate their Shopify store — turning catalogue expansion from a manual chore into an automated background process.