Exposing Text-Image Inconsistency Using Diffusion Models

1State University of New York at Buffalo

In the battle against widespread online misinformation, a growing problem is text-image inconsistency, where images are misleadingly paired with texts with different intent or meaning. Existing classification-based methods for text-image inconsistency can identify contextual inconsistencies but fail to provide explainable justifications for their decisions that humans can understand. Although more nuanced, human evaluation is impractical at scale and susceptible to errors. To address these limitations, this study introduces D-TIIL (Diffusion-based Text-Image Inconsistency Localization), which employs text-to-image diffusion models to localize semantic inconsistencies in text and image pairs. These models, trained on largescale datasets act as “omniscient” agents that filter out irrelevant information and incorporate background knowledge to identify inconsistencies. In addition, D-TIIL uses text embeddings and modified image regions to visualize these inconsistencies. To evaluate D-TIIL’s efficacy, we introduce a new TIIL dataset containing 14K consistent and inconsistent text-image pairs. Unlike existing datasets, TIIL enables assessment at the level of individual words and image regions and is carefully designed to represent various inconsistencies. D-TIIL offers a scalable and evidence-based approach to identifying and localizing text-image inconsistency, providing a robust framework for future research combating misinformation.


Interpolate start reference image.

There are 4 steps in our proposed pipeline:
1. Align text embedding to the input image by optimizing embedding
2. Generate mask from the difference of noise estimation and then conduct mask/text-guided image editing
3. Align text embedding to edited image to denoise the input text
4. Localize Inconsistency in the image and text



      title={Exposing Text-Image Inconsistency Using Diffusion Models},
      author={Mingzhen Huang and Shan Jia and Zhou Zhou and Yan Ju and Jialing Cai and Siwei Lyu},
      booktitle={International Conference on Learning Representations},