Localizing Knowledge in Diffusion Transformers

TL;DR

This paper introduces a method to localize and interpret knowledge within Diffusion Transformer models, enabling targeted model editing for personalization and unlearning with improved efficiency and minimal interference.

Contribution

It presents a novel, model- and knowledge-agnostic approach to localize knowledge in DiT models, facilitating interpretable analysis and efficient targeted updates.

Findings

Localized blocks are interpretable and causally linked to knowledge expression.

The method enables efficient, targeted fine-tuning for personalization and unlearning.

Improves model editing with minimal impact on unrelated content.

Abstract

Understanding how knowledge is distributed across the layers of generative models is crucial for improving interpretability, controllability, and adaptation. While prior work has explored knowledge localization in UNet-based architectures, Diffusion Transformer (DiT)-based models remain underexplored in this context. In this paper, we propose a model- and knowledge-agnostic method to localize where specific types of knowledge are encoded within the DiT blocks. We evaluate our method on state-of-the-art DiT-based models, including PixArt-alpha, FLUX, and SANA, across six diverse knowledge categories. We show that the identified blocks are both interpretable and causally linked to the expression of knowledge in generated outputs. Building on these insights, we apply our localization framework to two key applications: model personalization and knowledge unlearning. In both settings, our localized fine-tuning approach enables efficient and targeted updates, reducing computational cost, improving task-specific performance, and better preserving general model behavior with minimal interference to unrelated or surrounding content. Overall, our findings offer new insights into the internal structure of DiTs and introduce a practical pathway for more interpretable, efficient, and controllable model editing.