Where Culture Fades: Revealing the Cultural Gap in Text-to-Image Generation

TL;DR

This paper investigates the cultural biases in multilingual text-to-image models, revealing that they often produce culturally neutral or English-biased images, and proposes methods to enhance cross-cultural consistency.

Contribution

It identifies the root cause of cultural bias as insufficient activation of culture-related neurons and introduces alignment strategies to improve cultural fidelity in generated images.

Findings

Models often produce culturally neutral or biased images.

Proposed methods improve cultural consistency without sacrificing image quality.

Localization of culture-sensitive signals to specific neurons enables targeted enhancement.

Abstract

Multilingual text-to-image (T2I) models have advanced rapidly in terms of visual realism and semantic alignment, and are now widely utilized. Yet outputs vary across cultural contexts: because language carries cultural connotations, images synthesized from multilingual prompts should preserve cross-lingual cultural consistency. We conduct a comprehensive analysis showing that current T2I models often produce culturally neutral or English-biased results under multilingual prompts. Analyses of two representative models indicate that the issue stems not from missing cultural knowledge but from insufficient activation of culture-related representations. We propose a probing method that localizes culture-sensitive signals to a small set of neurons in a few fixed layers. Guided by this finding, we introduce two complementary alignment strategies: (1) inference-time cultural activation that amplifies the identified neurons without backbone fine-tuned; and (2) layer-targeted cultural enhancement that updates only culturally relevant layers. Experiments on our CultureBench demonstrate consistent improvements over strong baselines in cultural consistency while preserving fidelity and diversity.