T2I models generate traditional attire in 28.4% of images across 206 nationalities and five activities, with higher rates in Middle East/North Africa and Sub-Saharan Africa, stronger image-prompt alignment when country names are present, and frequent insertion of the word 'traditional' by one model.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Text-to-Image Models and Their Representation of People from Different Nationalities Engaging in Activities
T2I models generate traditional attire in 28.4% of images across 206 nationalities and five activities, with higher rates in Middle East/North Africa and Sub-Saharan Africa, stronger image-prompt alignment when country names are present, and frequent insertion of the word 'traditional' by one model.