Year: 2025
Type: article
Abstract: The field of AI-driven art and content generation has witnessed significant advancements. However, existing models typically operate as monolithic systems, lacking the nuanced colaboration, cross-domain capabilities, and iterative refinement necessary to meet complex user needs. This paper introduces the Multi-Agent System for Collaborative creation (MASC) framework, a novel approach integrating the collaborative, routing, and evaluative strengths of Multi-Agent Systems (MAS) with state-of-the-art generative models, such as Denoising Diffusion Probabilistic Models (DDPMs) and Diffusion Transformers (DiTs) for image synthesis. MASC leverages a team of specialized agents/modules (Task Agent, Domain Analysis Agent, Deepthink Module, Generator Module, Reflection Module) interacting through defined protocols and iterative loops. This enables the system to understand user intent, enrich creative prompts, generate multi-domain content (text, images, video, etc.), and perform evaluation-driven optimization. We elaborate on the MASC architecture, component roles, communication mechanisms, and its integration with core generative processes. Experimental results suggest MASC holds advantages in enhancing the detail richness and user intent alignment of generated content, opening new avenues for exploring more complex, cross-domain, and conceptually driven AI content creation.
Source: Frontiers in Art Research
Author: <p>Yi Luo</p>
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4413061965