How one studio unified 50+ AI models for image, video, and audio creation

A creative studio called Picturesque was built to consolidate over 50 AI models — spanning image generation, video, music, audio, upscaling, and motion control — under a single credit system and interface. The core engineering challenge was normalizing wildly different provider APIs, schemas, and pricing rules into one coherent product rather than a loose collection of integrations. The team built three entry points on a shared backend: a direct Studio interface, a node-based Workflow canvas, and an AI agent called Director that plans and executes multi-step creative tasks autonomously. Key lessons learned included the importance of instant credit refunds on provider failures, the value of a no-card free tier to drive adoption, and the ongoing maintenance burden of keeping model configurations current as new AI models ship continuously. The stack was kept deliberately simple — React, Node, Supabase, and Socket.IO — while a major backend refactor broke a 7,700-line monolithic file into structured routes and services.