Adaptive Clustering and Step-Size Optimization in Collaborative Distributed Diffusion-Based AIGC: Balancing Performance and Resource Utilization

This paper proposes a novel cloud-edge collaborative distributed diffusion model for AI-generated content (AIGC) such as image generation, which integrates adaptive clustering techniques with dynamic step-size optimization. The proposed model addresses the challenges of heterogeneous edge devices in real-world deployments. Experimental results demonstrate significant improvements in performance and efficiency with a 38.8% reduction in average generation time and a 15.6% increase in image quality (evaluate via CLIP score). The system shows enhanced resource utilization, improving cloud and edge utilization by 16.1 % and 36.6%, respectively. This research contributes to the advancement of collaborative distributed diffusion model, offering a scalable and adaptive framework for efficient AIGC services in dynamic environments along with potential applications extending to other computationally intensive tasks in cloud-edge systems.