Reproducibility Study of "ITI-GEN: Inclusive Text-to-Image Generation"

TL;DR

This paper reproduces and verifies the claims of ITI-GEN, a model designed to improve inclusiveness in text-to-image generation, and proposes enhancements to address its limitations.

Contribution

It confirms ITI-GEN's effectiveness and introduces Hard Prompt Search with negative prompting to mitigate its issues, especially with attribute disentanglement and scalability.

Findings

ITI-GEN improves diversity and quality of generated images

Hard Prompt Search with negative prompting enhances negation handling

Scaling to multiple attributes remains computationally challenging

Abstract

Text-to-image generative models often present issues regarding fairness with respect to certain sensitive attributes, such as gender or skin tone. This study aims to reproduce the results presented in "ITI-GEN: Inclusive Text-to-Image Generation" by Zhang et al. (2023a), which introduces a model to improve inclusiveness in these kinds of models. We show that most of the claims made by the authors about ITI-GEN hold: it improves the diversity and quality of generated images, it is scalable to different domains, it has plug-and-play capabilities, and it is efficient from a computational point of view. However, ITI-GEN sometimes uses undesired attributes as proxy features and it is unable to disentangle some pairs of (correlated) attributes such as gender and baldness. In addition, when the number of considered attributes increases, the training time grows exponentially and ITI-GEN struggles to generate inclusive images for all elements in the joint distribution. To solve these issues, we propose using Hard Prompt Search with negative prompting, a method that does not require training and that handles negation better than vanilla Hard Prompt Search. Nonetheless, Hard Prompt Search (with or without negative prompting) cannot be used for continuous attributes that are hard to express in natural language, an area where ITI-GEN excels as it is guided by images during training. Finally, we propose combining ITI-GEN and Hard Prompt Search with negative prompting.