As Metaverse emerges as the next-generation Internet paradigm, the ability to efficiently generate content is paramount. AI-Generated Content (AIGC) offers a promising solution to this challenge. However, the training and deployment of large AI models necessitate significant resources. To address this issue, we introduce an AIGC-as-a-Service (AaaS) architecture, which deploys AIGC models in wireless edge networks, ensuring ubiquitous access to AIGC services for Metaverse users. Nonetheless, a key aspect of providing personalized user experiences requires the careful selection of AIGC service providers (ASPs) capable of effectively executing user tasks. This selection process is complicated by environmental uncertainty and variability, a challenge not yet addressed well in existing literature. Therefore, we first propose a diffusion model-based AI-generated optimal decision (AGOD) algorithm, which can generate the optimal ASP selection decisions. We then apply AGOD to deep reinforcement learning (DRL), resulting in the Deep Diffusion Soft Actor-Critic (D2SAC) algorithm, which achieves efficient and effective ASP selection. Our comprehensive experiments demonstrate that D2SAC outperforms seven leading DRL algorithms. Furthermore, the proposed AGOD algorithm has the potential for extension to various optimization problems in wireless networks, positioning it a promising approach for the future research on AIGC-driven services in Metaverse. The implementation of our proposed method is available at: https://github.com/Lizonghang/AGOD.

翻译：随着元宇宙作为下一代互联网范式的兴起，高效生成内容的能力变得至关重要。AI生成内容（AIGC）为解决这一挑战提供了富有前景的方案。然而，大型AI模型的训练与部署需要大量资源。为此，我们提出一种AIGC即服务（AaaS）架构，将AIGC模型部署在无线边缘网络中，确保元宇宙用户能够泛在访问AIGC服务。然而，提供个性化用户体验的关键在于审慎选择能够有效执行用户任务的AIGC服务提供商（ASP）。这一选择过程因环境不确定性和动态变化而复杂化，现有文献尚未充分解决该挑战。因此，我们首先提出一种基于扩散模型的AI生成最优决策（AGOD）算法，该算法能够生成最优的ASP选择决策。随后我们将AGOD应用于深度强化学习（DRL），提出了深度扩散软演员-评论家（D2SAC）算法，实现了高效且有效的ASP选择。综合实验表明，D2SAC的性能优于七种主流的DRL算法。此外，所提出的AGOD算法具有扩展到无线网络中各类优化问题的潜力，有望成为未来元宇宙中AIGC驱动服务研究的重要方法。所提方法的实现代码已开源：https://github.com/Lizonghang/AGOD。