Vidster.ai SaaS is a web product for video translation, subtitle generation, and AI dubbing. It keeps the original Vidster.ai processing pipeline, but wraps it with the pieces a SaaS product needs: authentication, uploads, durable jobs, progress tracking, billing credits, and protected artifact downloads.
Why Use A Worker Architecture
Video localization jobs are long-running GPU tasks. They should not run inside the web request lifecycle. Vidster.ai SaaS uses a Postgres-backed queue and independent workers so the web app can stay responsive while GPU processes handle the heavy work.
The worker contract is simple:
- claim one queued job
- heartbeat while processing
- stop if cancellation is requested
- upload artifacts
- mark the job succeeded, failed, retried, or canceled
Local Development
The local stack uses Docker Compose for Postgres, local filesystem storage for uploads and artifacts, and one worker process per GPU. This matches the production model closely enough to test queue behavior before moving workers to RunPod or another GPU platform.
Production Direction
The next production step is to validate real Vidster.ai processing with the required WhisperX, spaCy, LLM, ASR, and TTS configuration. Remote workers can already use API-based input and artifact transfer; high-volume deployments can later move storage from local disk to S3 or R2 without changing the user workflow.