# Localizing and Editing Knowledge in Text-to-Image Generative Models

arXiv: https://arxiv.org/abs/2310.13730 · HTML: https://mlsanity.com/paper/2310.13730/

## Key facts

- Cited by 54 papers in the map (data through 14 Oct 2026).
- Placed in Image generation, text-to-image and editing, one of 14 directions in Diffusion models.
- First posted on arXiv 20 Oct 2023.

## Placed in

- Diffusion models / [Image generation, text-to-image and editing](https://mlsanity.com/diffusion/image-generation-text-to-image-and-editing/)

## Built on by (most cited first, 30 of 54)

- [MACE: Mass Concept Erasure in Diffusion Models](https://mlsanity.com/paper/2403.06135/) (2024)
- [Defensive Unlearning with Adversarial Training for Robust Concept Erasure in Diffusion Models](https://mlsanity.com/paper/2405.15234/) (2024)
- [SAeUron: Interpretable Concept Unlearning in Diffusion Models with Sparse Autoencoders](https://mlsanity.com/paper/2501.18052/) (2025)
- [A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models](https://mlsanity.com/paper/2502.17516/) (2025)
- [Erasing Concepts from Text-to-Image Diffusion Models with Few-shot Unlearning](https://mlsanity.com/paper/2405.07288/) (2024)
- [On Mechanistic Knowledge Localization in Text-to-Image Generative Models](https://mlsanity.com/paper/2405.01008/) (2024)
- [Decomposing and Editing Predictions by Modeling Model Computation](https://mlsanity.com/paper/2404.11534/) (2024)
- [Targeted Unlearning with Single Layer Unlearning Gradient](https://mlsanity.com/paper/2407.11867/) (2024)
- [TIDE : Temporal-Aware Sparse Autoencoders for Interpretable Diffusion Transformers in Image Generation](https://mlsanity.com/paper/2503.07050/) (2025)
- [TRCE: Towards Reliable Malicious Concept Erasure in Text-to-Image Diffusion Models](https://mlsanity.com/paper/2503.07389/) (2025)
- [Espresso: Robust Concept Filtering in Text-to-Image Models](https://mlsanity.com/paper/2404.19227/) (2024)
- [Padding Tone: A Mechanistic Analysis of Padding Tokens in T2I Models](https://mlsanity.com/paper/2501.06751/) (2025)
- [Detect-and-Guide: Self-regulation of Diffusion Models for Safe Text-to-Image Generation via Guideline Token Optimization](https://mlsanity.com/paper/2503.15197/) (2025)
- [Unveiling Concept Attribution in Diffusion Models](https://mlsanity.com/paper/2412.02542/) (2024)
- [SAEmnesia: Erasing Concepts in Diffusion Models with Supervised Sparse Autoencoders](https://mlsanity.com/paper/2509.21379/) (2025)
- [Rethinking Artistic Copyright Infringements in the Era of Text-to-Image Generative Models](https://mlsanity.com/paper/2404.08030/) (2024)
- [Mitigating Sexual Content Generation via Embedding Distortion in Text-conditioned Diffusion Models](https://mlsanity.com/paper/2501.18877/) (2025)
- [Safe and Reliable Diffusion Models via Subspace Projection](https://mlsanity.com/paper/2503.16835/) (2025)
- [Localizing Knowledge in Diffusion Transformers](https://mlsanity.com/paper/2505.18832/) (2025)
- [DLM-Scope: Mechanistic Interpretability of Diffusion Language Models via Sparse Autoencoders](https://mlsanity.com/paper/2602.05859/) (2026)
- [Joint Diffusion models in Continual Learning](https://mlsanity.com/paper/2411.08224/) (2024)
- [LoopAnimate: Loopable Salient Object Animation](https://mlsanity.com/paper/2404.09172/) (2024)
- [Precise Parameter Localization for Textual Generation in Diffusion Models](https://mlsanity.com/paper/2502.09935/) (2025)
- [Survey of User Interface Design and Interaction Techniques in Generative AI Applications](https://mlsanity.com/paper/2410.22370/) (2024)
- [Generative Model Unlearning: A Survey through Target Events, Unlearning Operators, and Evaluation Protocols](https://mlsanity.com/paper/2507.19894/) (2025)
- [TADA! Tuning Audio Diffusion Models through Activation Steering](https://mlsanity.com/paper/2602.11910/) (2026)
- [Low-Rank Continual Personalization of Diffusion Models](https://mlsanity.com/paper/2410.04891/) (2024)
- [SafeRedir: Prompt Embedding Redirection for Robust Unlearning in Image Generation Models](https://mlsanity.com/paper/2601.08623/) (2026)
- [Localized Concept Erasure in Text-to-Image Diffusion Models via High-Level Representation Misdirection](https://mlsanity.com/paper/2602.19631/) (2026)
- [REDEditing: Relationship-Driven Precise Backdoor Poisoning on Text-to-Image Diffusion Models](https://mlsanity.com/paper/2504.14554/) (2025)

## Abstract (arXiv)

Text-to-Image Diffusion Models such as Stable-Diffusion and Imagen have achieved unprecedented quality of photorealism with state-of-the-art FID scores on MS-COCO and other generation benchmarks. Given a caption, image generation requires fine-grained knowledge about attributes such as object structure, style, and viewpoint amongst others. Where does this information reside in text-to-image generative models? In our paper, we tackle this question and understand how knowledge corresponding to distinct visual attributes is stored in large-scale text-to-image diffusion models. We adapt Causal Mediation Analysis for text-to-image models and trace knowledge about distinct visual attributes to various (causal) components in the (i) UNet and (ii) text-encoder of the diffusion model. In particular, we show that unlike generative large-language models, knowledge about different attributes is not localized in isolated components, but is instead distributed amongst a set of components in the conditional UNet. These sets of components are often distinct for different visual attributes. Remarkably, we find that the CLIP text-encoder in public text-to-image models such as Stable-Diffusion contains only one causal state across different visual attributes, and this is the first self-attention layer corresponding to the last subject token of the attribute in the caption. This is in stark contrast to the causal states in other language models which are often the mid-MLP layers. Based on this observation of only one causal state in the text-encoder, we introduce a fast, data-free model editing method Diff-QuickFix which can effectively edit concepts in text-to-image models. DiffQuickFix can edit (ablate) concepts in under a second with a closed-form update, providing a significant 1000x speedup and comparable editing performance to existing fine-tuning based editing methods.