CMAG: Concept-Scaffolded Retrieval for Marketplace Avatar Generation
Researchers have developed CMAG, a framework designed to improve avatar generation in metaverse platforms by addressing the ambiguities and inconsistencies inherent in text-based retrieval. CMAG synthesizes an intermediate 3D concept scaffold to provide spatial and stylistic context, disambiguating user intent beyond simple text prompts. The system then uses a view-aware part discovery module and a prompt-conditioned taxonomy router to ensure category coverage and resolve semantic mismatches before a hybrid retriever assembles the final avatar from catalog assets, ensuring stylistic consistency and topological correctness. AI
IMPACT Introduces a novel approach to avatar generation that could improve user experience and asset consistency in metaverse platforms.