Towards Evaluating Robustness of Prompt Adherence in Text to Image Models

TL;DR

This paper develops a framework to evaluate the robustness of Text-to-Image models in adhering to prompts, revealing their struggles with simple variations and distribution conformity, and introduces a novel dataset and evaluation pipeline.

Contribution

It introduces a comprehensive evaluation framework, a novel dataset, and a pipeline for assessing prompt adherence and robustness in Text-to-Image models.

Findings

Models struggle with simple binary images and factors of variation.

Pre-trained VAEs show failure to generate images matching dataset distribution.

Evaluation pipeline effectively measures prompt adherence and robustness.

Abstract

The advancements in the domain of LLMs in recent years have surprised many, showcasing their remarkable capabilities and diverse applications. Their potential applications in various real-world scenarios have led to significant research on their reliability and effectiveness. On the other hand, multimodal LLMs and Text-to-Image models have only recently gained prominence, especially when compared to text-only LLMs. Their reliability remains constrained due to insufficient research on assessing their performance and robustness. This paper aims to establish a comprehensive evaluation framework for Text-to-Image models, concentrating particularly on their adherence to prompts. We created a novel dataset that aimed to assess the robustness of these models in generating images that conform to the specified factors of variation in the input text prompts. Our evaluation studies present findings on three variants of Stable Diffusion models: Stable Diffusion 3 Medium, Stable Diffusion 3.5 Large, and Stable Diffusion 3.5 Large Turbo, and two variants of Janus models: Janus Pro 1B and Janus Pro 7B. We introduce a pipeline that leverages text descriptions generated by the gpt-4o model for our ground-truth images, which are then used to generate artificial images by passing these descriptions to the Text-to-Image models. We then pass these generated images again through gpt-4o using the same system prompt and compare the variation between the two descriptions. Our results reveal that these models struggle to create simple binary images with only two factors of variation: a simple geometric shape and its location. We also show, using pre-trained VAEs on our dataset, that they fail to generate images that follow our input dataset distribution.