MAVEN A Multi-Agent Framework for Multicultural Text-to-Video Generation

2026-05-15

AI Summary

Existing text-to-video generation models struggle to accurately represent diverse cultural characteristics within a single prompt. To address this limitation, this work proposes MAVEN, a multi-agent framework that decouples prompts into three distinct dimensions—person, action, and location—and employs specialized agents operating in parallel or sequentially to collaboratively optimize cultural fidelity in both monocultural and cross-cultural scenarios. We introduce the first multi-agent prompt optimization mechanism tailored for multicultural representation and construct the first text-to-video evaluation benchmark encompassing Chinese, American, and Romanian cultures. Experimental results demonstrate that the parallel specialization strategy significantly enhances cultural relevance while preserving high visual quality and temporal consistency.

Abstract

Text-to-video (T2V) generation has rapidly progressed in visual fidelity, yet its ability to faithfully represent multiple cultures within a single prompt remains underexplored. We introduce MAVEN, a multi-agent prompt refinement framework designed to improve cultural fidelity in both mono-cultural and cross-cultural T2V generation. MAVEN decomposes prompts into person, action, and location dimensions, handled by specialized agents operating in parallel or sequentially. To support systematic evaluation, we contribute a new benchmark of 243 culturally grounded prompts and 972 corresponding videos, spanning three cultures (Chinese, American, Romanian), three action categories, and both mono-cultural and cross-cultural scenarios. Evaluations combining CLIP-based metrics, VLM-as-judge assessments, and videoquality measures show that multi-agent refinement, particularly parallel specialization, significantly improves cultural relevance while preserving visual quality and temporal consistency. The dataset and code are available athttps://github.com/AIM-SCU/CRAFT

Problem

• text-to-video generation
• cultural representation
• multicultural fidelity
• prompt engineering
• cross-cultural generation

Innovation

• multi-agent framework
• cultural fidelity
• text-to-video generation
• prompt refinement
• cross-cultural representation