On Mechanistic Knowledge Localization in Text-to-Image Generative Models

TL;DR

This paper introduces a new method for localizing visual attribute knowledge within text-to-image models, enabling more efficient and precise model editing by identifying specific layers responsible for different visual features.

Contribution

The paper proposes the concept of Mechanistic Localization and introduces LocoGen and LocoEdit for effective knowledge localization and model editing in recent text-to-image models.

Findings

Localization of visual attributes to specific UNet layers

LocoEdit enables fast, closed-form model editing

Improved understanding of model knowledge distribution

Abstract

Identifying layers within text-to-image models which control visual attributes can facilitate efficient model editing through closed-form updates. Recent work, leveraging causal tracing show that early Stable-Diffusion variants confine knowledge primarily to the first layer of the CLIP text-encoder, while it diffuses throughout the UNet.Extending this framework, we observe that for recent models (e.g., SD-XL, DeepFloyd), causal tracing fails in pinpointing localized knowledge, highlighting challenges in model editing. To address this issue, we introduce the concept of Mechanistic Localization in text-to-image models, where knowledge about various visual attributes (e.g., "style", "objects", "facts") can be mechanistically localized to a small fraction of layers in the UNet, thus facilitating efficient model editing. We localize knowledge using our method LocoGen which measures the direct effect of intermediate layers to output generation by performing interventions in the cross-attention layers of the UNet. We then employ LocoEdit, a fast closed-form editing method across popular open-source text-to-image models (including the latest SD-XL)and explore the possibilities of neuron-level model editing. Using Mechanistic Localization, our work offers a better view of successes and failures in localization-based text-to-image model editing. Code will be available at https://github.com/samyadeepbasu/LocoGen.