Date of Award
11-2025
Document Type
Dissertation
Publisher
Santa Clara : Santa Clara University, 2026
Degree Name
Master of Science (MS)
Department
Computer Science and Engineering
First Advisor
Oana Ignat
Abstract
We present a novel approach to cross-cultural meme transcreation using vision-language models (VLMs), focusing on the preservation of communicative intent while adapting cultural references for target audiences. Our three-stage hybrid transcreation pipeline leverages a specialized VLM, LLaVA 1.6, for analysis and caption generation, in conjunction with FLUX.1 Schnell model for culturally-adapted visual templates, to transform memes across Chinese and American cultural contexts. Unlike traditional translation approaches, our method strategically preserves universal meme formats while replacing culture-specific elements, maintaining humor and relatability. Through a comprehensive evaluation involving both human assessors and automated metrics, we demonstrate that prompt-engineered VLMs with optimized parameters performed well in cultural authenticity and intent preservation. Our contributions include: (1) a systematic framework for hybrid meme transcreation, (2) bidirectional evaluation across four diverse cultural contexts, and (3) novel insights into the universal versus culture-specific expression patterns in multimodal humor. Our dataset and code are available at: https://github.com/AIM-SCU/CultuR-AI-ze . We hope these resources facilitate further research into the complex interplay of visual symbolism and linguistic humor in generative AI.
Recommended Citation
Zhao, Yuming, "Bridging Cultural Gaps with Vision-Language Models for Automated Meme Transcreation" (2025). Computer Science and Engineering Master's Theses. 61.
https://scholarcommons.scu.edu/cseng_mstr/61
Comments
This work is licensed under a Creative Commons “Attribution-NonCommercial 4.0 International” license.