Product Photo Standardizer
Gets thousands of product photos ready for ecommerce, respecting the brand manuals, with AI that understands images.
The problem
Editing backgrounds, dimensions, classification and ordering for thousands of product photos was fully manual.
What I did
I worked on the standardization pipeline, adding duplicate-image detection and the usage guide: a flow of AI cutout and shadows, GPT-4o vision classification and RAG over the brand manuals in PDF.
The outcome
In production on the portal, adopted by the ecommerce area.
From raw photo to listing
- Raw product photoInput
- Duplicate detectionDo not process the same twice
- AI cropping and shadowsBackground and dimensions
- Vision classificationGPT-4o
- Validation against the manualsRAG over the brand PDFs
- Photo ready for ecommerceOutput
The criteria do not come from the model: they come from the brand manuals, which live in PDFs and enter the pipeline through RAG. The AI crops, classifies and checks against that standard instead of deciding on its own what looks right.
Stack
- Next.js 16
- React 19
- TypeScript
- GPT-4o
- Gemini
- Image-editing API
- Pinecone
- Supabase
- Docker
Blue Star Group internal project — the code is not public.
Next project: Partner FOB Cost API →