Cost-Aware Routing for Efficient Text-To-Image Generation
Researchers have developed a cost-aware routing system for text-to-image generation that dynamically adjusts computational resources based on prompt complexity. This framework routes each prompt to the most suitable generation function, which could involve varying the number of denoising steps in a diffusion model or selecting an entirely different model. By learning to reserve intensive computations for complex prompts and using more economical options for simpler ones, the system aims to optimize the trade-off between image quality and computational cost. Experiments on COCO and DiffusionDB datasets showed that this routing approach, utilizing nine pre-trained models, achieved higher average quality than any single model could alone. AI
IMPACT This approach could lead to more efficient and cost-effective image generation by dynamically allocating computational resources based on prompt complexity.