The Latent Color Subspace: Emergent Order in High-Dimensional Chaos

2026/03/12 by Mateusz Pach, Jessica Bader, Quentin Bouniot +2 · 1 voice
Computer Science · #cs.AI #cs.CV #cs.LG

paper · pdf · doi:10.48550/arxiv.2603.12261

arxiv published 2026/03/12 · arxiv updated 2026/06/10

Abstract

Text-to-image generation models have advanced rapidly, yet achieving fine-grained control over generated images remains difficult, largely due to limited understanding of how semantic information is encoded. We develop an interpretation of the color representation in the Variational Autoencoder latent space of FLUX.1 [Dev], revealing a structure reflecting Hue, Saturation, and Lightness. We verify our Latent Color Subspace (LCS) interpretation by demonstrating that it can both predict and explicitly control color, introducing a fully training-free method in FLUX based solely on closed-form latent-space manipulation. Code is available at https://github.com/ExplainableML/LCS.

Citations

• Color encoding in Latent Space of Stable Diffusion Models
• Stitch: Training-Free Position Control in Multimodal Diffusion Transformers
• Seg4Diff: Unveiling Open-Vocabulary Segmentation in Text-to-Image Diffusion Transformers
• Controllable-Continuous Color Editing in Diffusion Model via Color Mapping
• Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Models
• Stable Diffusion Models are Secretly Good at Visual In-Context Learning
• Story2Board: A Training-Free Approach for Expressive Storyboard Generation
• Qwen-Image Technical Report
• SUB: Benchmarking CBM Generalization via Synthetic Attribute Substitutions
• FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Space
• Interpreting Large Text-to-Image Diffusion Models with Dictionary Learning
• CDST: Color Disentangled Style Transfer for Universal Style Reference Customization
• Does Feasibility Matter? Understanding the Impact of Feasibility on Synthetic Training Data
• Generative AI for Urban Planning: Synthesizing Satellite Imagery via Diffusion Models
• Dissecting and Mitigating Diffusion Bias via Mechanistic Interpretability
• DCEdit: Dual-Level Controlled Image Editing via Precisely Localized Semantics
• Leveraging Semantic Attribute Binding for Free-Lunch Color Control in Diffusion Models
• Color Alignment in Diffusion
• ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features
• Towards a Mechanistic Explanation of Diffusion Model Generalization
• Revelio: Interpreting and leveraging semantic information in diffusion models
• ColorPeel: Color Prompt Learning with Diffusion Models via Color and Shape Disentanglement
• ReNO: Enhancing One-step Text-to-Image Models through Reward-based Noise Optimization
• Scaling Rectified Flow Transformers for High-Resolution Image Synthesis
• Control Color: Multimodal Diffusion-based Interactive Image Colorization
• GenEval: An Object-Focused Framework for Evaluating Text-to-Image Alignment
• PixArt-α: Fast Training of Diffusion Transformer for Photorealistic Text-to-Image Synthesis
• PathLDM: Text conditioned Latent Diffusion Model for Histopathology
• IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Models
• Dequantization and Color Transfer with Diffusion Models
• SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis
• SceneGenie: Scene Graph Guided Diffusion Models for Image Synthesis
• DINOv2: Learning Robust Visual Features without Supervision
• Adding Conditional Control to Text-to-Image Diffusion Models
• Adding Conditional Control to Text-to-Image Diffusion Models
• Attend-and-Excite: Attention-Based Semantic Guidance for Text-to-Image Diffusion Models
• Scalable Diffusion Models with Transformers
• What the DAAM: Interpreting Stable Diffusion Using Cross Attention
• Flow Matching for Generative Modeling
• Building Normalizing Flows with Stochastic Interpolants
• Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow
• Prompt-to-Prompt Image Editing with Cross Attention Control
• High-Resolution Image Synthesis with Latent Diffusion Models
• High-Resolution Image Synthesis with Latent Diffusion Models
• The Unreasonable Effectiveness of Deep Features as a Perceptual Metric
• Image quality assessment: from error visibility to structural similarity

Discussions

• 5/ The Latent Color Subspace: Emergent Order in High-Dimensional Chaos Mateusz Pach, Jessica Bader, @qbouniot.bsky.social, Serge Belongie, @zeynepakata.bsky.social [Paper]: arxiv.org/abs/2603.12261 [C [bsky, 0 points, 1 comments]

Related

• Color Bind: Exploring Color Perception in Text-to-Image Models
• Color encoding in Latent Space of Stable Diffusion Models
• LASERS: LAtent Space Encoding for Representations with Sparsity for Generative Modeling
• Trained Latent Space Navigation to Prevent Lack of Photorealism in Generated Images on Style-based Models
• Semantic-driven Colorization