A new study published on arXiv investigates social biases in text-to-image generation models, specifically comparing bias expression across photo, storyboard, and comic generation tasks. The research adapted the BBG framework to evaluate six text-to-image models, finding that proprietary models produced 25.9% biased outputs in photo generation. This bias increased to 35.5% for storyboards and 44.1% for comics, indicating that narrative visual formats reveal biases more explicitly through elements like character positioning and narrative resolution. AI
IMPACT Highlights the need for diverse evaluation methods for text-to-image models to address biases in narrative visual formats.
RANK_REASON Academic paper on AI model bias. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
