VisionFlow — AI Image & Video Platform

$ Overview

Took part in building a multimodal AIGC platform (text-to-image / text-to-video / workflow orchestration) from zero to launch, aimed at cross-border e-commerce: batch generation of product main images and detail pages. It now runs as a multi-domain, multi-brand SaaS (another brand: Terln).

$ Highlights

• E-commerce focus: batch generation of product main images and detail pages; official workflows out of the box, community workflows publishable
• Workflow orchestration: built on n8n, with a self-developed canvas
• Task scheduling: decoupled submission from execution with a distributed queue for tiered, prioritized dispatch
• Billing engine: a freeze-confirm-release balance pre-authorization model supporting batch billing and rollback on failure
• Model integration: multi-provider routing with circuit breakers; key rotation and result caching to keep costs down
• Engineering: admin monitoring dashboard, API specs and deploy docs, canary releases with fast rollback

$ Tech Stack

• Next.js
• TypeScript
• PostgreSQL
• Redis
• n8n