GenRouter: Unified Workflow Routing for Agentic Image Generation
The paper’s pitch is a router that sends each image prompt to the cheapest workflow that can handle it.
GenRouter standardizes agentic image-generation pipelines into shared primitives and executable templates, then chooses among them using demand profiling, experience matching, and Pareto filtering. The authors say this avoids pushing simple prompts through heavyweight fixed pipelines. In their experiments, it improved visual alignment while cutting execution costs by more than 95% and latency by 65% versus heavyweight static workflows. They also claim the system improves from accumulated experience, helping zero-shot generalization while halving computational overhead. HF Daily Papers' note
GenRouter standardizes agentic image-generation pipelines into shared primitives and executable templates, then chooses among them using demand profiling, experience matching, and Pareto filtering. The authors say this avoids pushing simple prompts through heavyweight fixed pipelines. In their experiments, it improved visual alignment while cutting execution costs by more than 95% and latency by 65% versus heavyweight static workflows. They also claim the system improves from accumulated experience, helping zero-shot generalization while halving computational overhead. HF Daily Papers' note
score 5