Seed Audio Prompts

 Seed Audio Prompts is a full-stack AI media generation platform transforming text prompts into audio, images, video, 3D models, and voice clones — all powered by a unified multi-provider AI backend.

     Live: seedaudioprompts.com · · GitHub: github.com/gladlyknow/seedaudioprompts     Architecture. The platform runs entirely on Cloudflare's edge: Next.js App Router deployed via OpenNext on Workers, Cloudflare Queues for async job processing, Hyperdrive connecting to a Supabase PostgreSQL database, and R2 for object storage. A provider abstraction layer     orchestrates five AI backends — Doubao/Volcengine, Fal.ai, Replicate, Gemini, and Kie — with credit-based pricing configurable per media type, model, and scene.▔      async reliability. Tasks enter a Cloudflare Queue where the consumer worker executes a two-phase credit flow: reserve credits upfront, call the AI provider, then confirm or release based on the result. Idempotency keys prevent duplicate charges on retry, and exponential backoff     handles transient failures. Async media types (video, 3D) complete via provider webhooks that persist ephemeral URLs to durable R2 storage before confirming credits.

     Product features. Full zh/en internationalization with browser-language detection. Integrated payments via Creem, Stripe, and PayPal with sandbox/production moderation split. File lifecycle management with configurable 30-day retention and automated R2 cleanup via cron. FIFO      credit expiry, server-side content moderation, and independent dev/prod deployments across separate PostgreSQL schemas. Built with Next.js 15, TypeScript, Drizzle ORM, Radix UI, and Tailwind CSS.