More about HKUST
Distributed Inference for Multi-model Generative AI Workflows in ComfyUI
The Hong Kong University of Science and Technology Department of Computer Science and Engineering Final Year Thesis Oral Defense Title: "Distributed Inference for Multi-model Generative AI Workflows in ComfyUI" by MORSI Mohamed Sobhy Mohamed Hassan Abstract: This thesis presents the design, implementation, and evaluation of a distributed inference system for ComfyUI, aimed at overcoming the computational constraints associated with running ComfyUI workflows on single machines and the privacy concerns associated with running them on cloud servers. Complex workflows that previously exceeded single-machine capabilities can now be executed efficiently through our distributed approach. This research contributes to the field of distributed systems for AI applications by providing an extensible framework that can scale with increasing model complexity and computational demands, making advanced AI workflows more accessible to users with limited individual computing resources. Date : 3 May 2025 (Satuarday) Time : 12:20 - 13:00 Venue : Room 2130B (near lift 19), HKUST Advisor : Prof. GUO Song 2nd Reader : Prof. WU Dekai