AI / Seller operationsPrivate single-seller system

From one phone photo to a complete product pack

A phone-friendly studio extracts visible facts, creates product-scene alternatives, checks likely visual changes, writes editable Thai copy, builds short vertical video, and keeps the results together.

Durable image and video queues · source-versus-result checks · Thai copy · shared gallery · credit use and refunds

The flow

How the pieces connect

  1. Phone photo
  2. Visible facts
  3. Image alternatives
  4. Fidelity check
  5. Editable Thai copy
  6. Short video and shared gallery
01

The job

A seller needed a repeatable content workflow, not a prompt box and a pile of downloaded files.

02

The hard part

Keep provider credentials and model controls away from the seller interface, make paid work durable, and prevent generated claims from outrunning visible product facts.

03

How I built it

  1. Extract visible facts before writing copy or generating a scene.
  2. Run images and video through durable jobs so closing the app does not lose paid work.
  3. Compare the source and generated product, then show warnings instead of claiming perfect fidelity.
  4. Keep model choice, prompts, credentials, credits, refunds, and provider controls on the owner side.
04

What I verified

  • Phone upload and camera capture
  • Structured fact extraction, image generation, fidelity comparison, and Thai copy
  • Durable image and video queues with usage deductions and refunds
  • Seller and owner roles, persistent settings, and shared history

Current state: The current build is for one seller. A multi-tenant version is not built, and source-versus-result checks cannot guarantee that generated media preserves every product detail.