Spec-to-Sell

Turn supplier specs into accurate, brand-consistent PoD listing infographics to include in your listings to sell.

Comment

Inspiration

Print-on-demand sellers often receive a loose set of supplier product photos and raw specifications, then need to turn them into a cohesive listing-image set. The work is repetitive, visually inconsistent, and risky when a seller accidentally makes a claim that the supplier never supported.

What it does

Spec-to-Sell turns supplier facts and vendor product photos into a suggested 2–6 slide infographic set for a product listing. The seller enters a lightweight shop brand profile, uploads vendor photos, and pastes supplier specifications.

The app creates a brand kit that locks the palette, typography, spacing, image treatment, and logo placement across every exported slide. It then builds an evidence ledger for supplier claims. Supported claims include their source evidence; unsupported or conflicting claims are held for seller review before they can be included.

The final listing images are exported as 2000 × 2000 PNGs. Vendor product photos are only placed, cropped, or scaled in deterministic layouts—Spec-to-Sell does not generate or alter the product image.

How we built it

The project is a Next.js and TypeScript web application using React/Konva for deterministic infographic rendering and JSZip for batch export.

Its server-side analysis route uses GPT-5.6 Terra through the OpenAI Responses API to process supplier text and vendor images as multimodal input. The model returns structured evidence, confidence, warnings, and an evidence-backed storyboard proposal. The app applies deterministic rules to keep the suggested set between two and six slides.

Codex accelerated the application scaffold, evidence-first data model, deterministic canvas exporter, test coverage, and production verification. GPT-5.6 Terra is used for supplier-fact extraction and storyboard planning, not for product-image generation.

Evidence policy

  • Supplier text and vendor images establish product truth.
  • The app requires cited evidence for supported claims.
  • Unsupported claims remain marked needs_review until the seller explicitly confirms them.
  • The renderer uses source-locked vendor photos and an approved brand system for every exported slide.

Try it

The repository includes a packaged vendor tote sample so judges can run the complete editor and export workflow without an API key. Set OPENAI_API_KEY in .env.local to use the live GPT-5.6 Terra analysis route with supplier inputs.

Built With

  • gpt-5.6-terra
  • jszip
  • next.js
  • openai-api
  • react
  • react-konva
  • tailwind-css
  • typescript
  • zod
Share this project:

Updates