ServerlessT2I: Efficient Text-to-Image Workflow Serving on a Serverless Platform

ServerlessT2I decomposes text-to-image workflows into loosely coupled model functions to enable independent per-model scaling and scheduling. This approach replaces monolithic GPU function deployments with a system that supports transparent GPU-resident communication and fine-grained fairness-aware scheduling in multi-tenant clusters.