Color encoding in Latent Space of Stable Diffusion Models

2025-12-10

AI Summary

This study investigates the encoding structure of perceptual attributes—color, brightness, and shape—in the latent space of Stable Diffusion. To isolate attribute-specific representations, we construct a controllable synthetic dataset and apply principal component analysis (PCA), channel-wise decomposition, and cosine similarity metrics. We find that color information is predominantly encoded along a circular, opponent-axis structure in the c₃/c₄ channels, whereas brightness and shape are concentrated in the c₁/c₂ channels. This is the first demonstration that Stable Diffusion’s latent space adheres to efficient coding principles, exhibiting geometric interpretability and approximate attribute disentanglement. The results establish a novel opponent-geometric representation paradigm for color within specific latent channels and provide a foundational framework for latent-space-based controllable image editing and model interpretability research.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recent advances in diffusion-based generative models have achieved remarkable visual fidelity, yet a detailed understanding of how specific perceptual attributes - such as color and shape - are internally represented remains limited. This work explores how color is encoded in a generative model through a systematic analysis of the latent representations in Stable Diffusion. Through controlled synthetic datasets, principal component analysis (PCA) and similarity metrics, we reveal that color information is encoded along circular, opponent axes predominantly captured in latent channels c_3 and c_4, whereas intensity and shape are primarily represented in channels c_1 and c_2. Our findings indicate that the latent space of Stable Diffusion exhibits an interpretable structure aligned with a efficient coding representation. These insights provide a foundation for future work in model understanding, editing applications, and the design of more disentangled generative frameworks.

Problem

• Explores color encoding in Stable Diffusion's latent space
• Reveals circular opponent axes in channels c_3 and c_4
• Identifies interpretable structure for model understanding and editing

Innovation

• Analyzes color encoding in Stable Diffusion latent space
• Uses PCA and similarity metrics on synthetic datasets
• Finds color encoded in circular axes in specific channels