DreamCast AI: A Provider-Agnostic, Job-Based Architecture for Project-Oriented Multimodal AI Content Creation

Abstract

Creating even a short multimedia piece currently requires several separate generative services for text, images, video, speech and music, each with its own account, quota, option set and failure behaviour. This paper describes DreamCast AI, a web application that organises these services into one project-oriented workflow, and analyses the engineering decisions behind it. The system combines a React and TypeScript frontend with a FastAPI backend, a relational database and an object-storage abstraction. Generation requests share a single two-phase pipeline: synchronous validation, administrator-defined defaults, plan-limit checking and usage reservation, followed by asynchronous execution of a persistent job through a provider adapter.