Discovering Divergent Representations between Text-to-Image Models
This work addresses the need to understand and compare generative model biases for researchers and practitioners in AI, though it is incremental as it builds on existing methods for model analysis.
The paper tackles the problem of discovering when and how visual representations diverge between text-to-image models, introducing CompCon, an evolutionary search algorithm that identifies visual attributes more prevalent in one model's output than another and links them to prompt concepts, with evaluation on a dataset of 60 input-dependent differences.
In this paper, we investigate when and how visual representations learned by two different generative models diverge. Given two text-to-image models, our goal is to discover visual attributes that appear in images generated by one model but not the other, along with the types of prompts that trigger these attribute differences. For example, "flames" might appear in one model's outputs when given prompts expressing strong emotions, while the other model does not produce this attribute given the same prompts. We introduce CompCon (Comparing Concepts), an evolutionary search algorithm that discovers visual attributes more prevalent in one model's output than the other, and uncovers the prompt concepts linked to these visual differences. To evaluate CompCon's ability to find diverging representations, we create an automated data generation pipeline to produce ID2, a dataset of 60 input-dependent differences, and compare our approach to several LLM- and VLM-powered baselines. Finally, we use CompCon to compare popular text-to-image models, finding divergent representations such as how PixArt depicts prompts mentioning loneliness with wet streets and Stable Diffusion 3.5 depicts African American people in media professions. Code at: https://github.com/adobe-research/CompCon