Development of a Federated Platform for Accessible AI-Assisted Photo Editing

Abstract

The rapid advancements in generative artificial intelligence have led to growing demand for accessible and open-source tools that enable users to create and edit images using AI models. However, the high computational requirements and limited variety of models offered by existing applications pose significant barriers to entry. This paper presents a novel federated system of open-source queue servers that connects user clients to GPU clients, allowing users to host, modify, and share AI models freely. The system incorporates a generalisation layer for handling various tasks, a federation protocol for forwarding user requests to appropriate queue servers, and a trust-based priority scheduling scheme for managing bad actors. Experimental results demonstrate the effectiveness of the proposed system in enhancing accessibility and efficiency in generative art. The developed federated network and queue servers have potential applications beyond photo editing, creating new possibilities for collaborative and decentralised AI-assisted content creation. The methodology for this study involves several stages of development to create a fully functioning federated platform for photo editing using AI models. The first stage focuses on the development of a queueing system, which is built using Python, Flask, and WebSockets for communication. The second stage will involve the creation of a client GPU server, which is built using Python, Flask, and SocketIO. In the third stage, a frontend photo editing application is developed using React.js. To ensure that the system can support multiple generative AI models, a Generalization Layer for job execution is created in the fourth stage. The fifth stage involves the development of a federated protocol for server-server communication. Strategies are implemented in the sixth stage to limit bad actors in the system.