Aligned but Stereotypical? The Hidden Influence of System Prompts on Social Bias in LVLM-Based Text-to-Image Models

TL;DR

The paper reveals that LVLM-based text-to-image models exhibit stronger demographic biases than non-LVLM systems, and that system prompts are a key mechanism driving this bias. To enable robust evaluation, the authors create a 1,024-prompt benchmark spanning four linguistic levels and measure bias with LVLM-based judges, showing a strong link between prompt complexity, demographic cues, and alignment. They provide a mechanistic analysis demonstrating that system prompts encode demographic priors that shift token probabilities and text embeddings, ultimately biasing image synthesis. As a practical contribution, they introduce FairPro, a training-free meta-prompting framework that self-audits prompts and replaces the system prompt with a fairness-aware version at test time, achieving substantial bias reduction while preserving text–image alignment across two LVLM-based T2I models. The work highlights the pivotal role of system prompts in bias propagation and offers a deployable method to build more socially responsible LVLM-based generative systems.

Abstract

Large vision-language model (LVLM) based text-to-image (T2I) systems have become the dominant paradigm in image generation, yet whether they amplify social biases remains insufficiently understood. In this paper, we show that LVLM-based models produce markedly more socially biased images than non-LVLM-based models. We introduce a 1,024 prompt benchmark spanning four levels of linguistic complexity and evaluate demographic bias across multiple attributes in a systematic manner. Our analysis identifies system prompts, the predefined instructions guiding LVLMs, as a primary driver of biased behavior. Through decoded intermediate representations, token-probability diagnostics, and embedding-association analyses, we reveal how system prompts encode demographic priors that propagate into image synthesis. To this end, we propose FairPro, a training-free meta-prompting framework that enables LVLMs to self-audit and construct fairness-aware system prompts at test time. Experiments on two LVLM-based T2I models, SANA and Qwen-Image, show that FairPro substantially reduces demographic bias while preserving text-image alignment. We believe our findings provide deeper insight into the central role of system prompts in bias propagation and offer a practical, deployable approach for building more socially responsible T2I systems.

Figures (7)

• Figure 1: Social bias in recent T2I models. Given the neutral prompt "A botanist," non-LVLM-based models (left) produce demographically diverse images, whereas LVLM-based models (middle) are biased toward specific gender and ethnic groups. Applying our FairPro (right) notably reduces these biases and yields more diverse generations while preserving text-image alignment.
• Figure 2: Social bias and alignment in LVLM-based vs. non-LVLM T2I models. We evaluate recent text-to-image models across three dimensions: overall demographic bias, bias variation under increasing prompt complexity, and text–image alignment. LVLM-based models consistently exhibit stronger social biases than non LVLM-based models. Furthermore, bias increases with prompt complexity and follows a trend similar to text–image alignment.
• Figure 3: Analyzing decoded prompts. Decoded texts reveal demographic assumptions introduced by system prompts, which correlate with biases in the final generated images.
• Figure 4: Impact of system prompts on linguistic bias. We conduct a controlled analysis to quantify the effect of system prompts on linguistic bias within LVLMs. Removing system prompts mitigates gender bias, as reflected in both (a) token probability distributions and (b) text representations.
• Figure 5: Qualitative comparison. While the default system prompt tends to produce demographically biased outputs, our proposed FairPro method generates individuals with greater diversity, even when explicit demographic attributes are specified. Furthermore, FairPro maintains demographic diversity and prompt coherence even under long and complex prompts. Best viewed zoomed in.