Abstract Artificial intelligence-generated image content (AIGIC) is produced through the extraction of features and patterns from a vast image dataset, requiring substantial computational resources for training. This study aims to enhance image processing and response time on terminal devices by utilizing edge computing technology to offload specific training tasks to edge nodes. Additionally, task offloading and resource allocation strategies are developed to effectively generate image content on terminal devices. Edge computing aims to execute computing tasks in close proximity to data sources; however, the computing resources of edge devices are limited. Therefore, the development of suitable resource allocation strategies for resource-constrained environments is crucial in edge computing research. Serverless computing, which heavily relies on container technology for program hosting, is recognized as one of the most suitable architectures for edge computing. WebAssembly (WASM) is a binary instruction format that operates on a stack and enables the execution of computing tasks on both client and server sides. Its advantages encompass reducing cold start time, enhancing efficiency, and improving portability, thereby addressing challenges encountered by container technology in Serverless deployments. This paper commences with an introduction and analysis of the research status of Serverless and WASM, subsequently delving into the investigation of task offloading and resource allocation in edge computing within the Serverless architecture supported by WASM. To facilitate collaboration among edge nodes, an enhanced deep reinforcement learning algorithm, called entropy-based Proximal Policy Optimization (E-PPO2), is employed. This algorithm allows edge devices to share a global reward and continuously update parameters, leading to an optimized response strategy and maximizing the utilization of edge device resources.