Catalogue Refinery
AI enrichment of the product catalogue (vision + colour + text, fused). Run batches to grind through the -product backlog. This is the pilot & retro-fix — measure accuracy and conflict rate here before charging any store.
style-ready
Schema v · products · worker
Each product ≈ one Gemini vision call (~$0.002, ~9s). Colour variants sharing an image are deduped to one call. Keep batches ≤10 so the request doesn't time out; if one does, the batch still finishes server-side — just refresh. Use auto-run to grind the whole backlog.
Last batch
Fields sent to review:
Re-refine
Flip rows back to pending to run them again — after a review clean-up, a schema/prompt version bump, or to reset.
Image standardisation
Sharp trim → white square canvas → resize (deterministic, free, no AI). Model shots are skipped. Background removal is a setting (needs the Cloudinary add-on). Try-on uses the standardised image when present.
Review queue
Nothing to review — every enriched product auto-confirmed. Conflicts (vision vs store) land here for a one-tap fix.