Deep Dreams Are Made of This: Visualizing Monosemantic Features in Diffusion Models

TL;DR

This paper introduces latent visualization by optimization (LVO), a technique extending feature visualization to latent diffusion models, enabling clear visualization of monosemantic features through disentangled autoencoder representations.

Contribution

The paper presents LVO, a novel interpretability method for diffusion models that disentangles features and visualizes monosemantic concepts in the latent space.

Findings

SAE features produce clear visualizations of recognizable concepts

Regularization techniques transfer from pixel-space to latent domain

LVO provides insights into feature activation mechanisms

Abstract

This paper proposes latent visualization by optimization (LVO), a mechanistic interpretability technique that extends feature visualization by optimization - originally developed for convolutional neural networks - to latent diffusion models. LVO employs sparse autoencoders (SAEs) to disentangle polysemantic layer representations into monosemantic features. Key contributions include latent-space optimization, time-step activity analysis, schedule-matched noise injection, prior initialization through feature steering, and suitable regularization strategies. We demonstrate the method on Stable Diffusion 1.5 fine-tuned on the Style50 dataset, showing that SAE features produce clear visualizations of recognizable concepts - including diagonal compositions, human figures, roses, cables, and waterfall foam - that correlate with dataset examples, while the baseline without disentanglement produces less coherent results. We further show that regularization techniques from pixel-space feature visualization transfer to the latent domain, though they require different configurations for the raw-layer and SAE variants. Compared to dataset examples and steering, LVO provides complementary insights by directly revealing what activates a feature rather than its downstream effects.