# Emergence and Evolution of Interpretable Concepts in Diffusion Models

arXiv: https://arxiv.org/abs/2504.15473 · HTML: https://mlsanity.com/paper/2504.15473/

## Key facts

- Cited by 22 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.
- Builds on 36 of its 48 references that are in the map.
- First posted on arXiv 21 Apr 2025.

## Placed in

- Diffusion models / [Image generation, text-to-image and editing](https://mlsanity.com/diffusion/image-generation-text-to-image-and-editing/)
- Mechanistic interpretability / [Interpreting and steering diffusion models](https://mlsanity.com/mechinterp/interpreting-and-steering-diffusion-models/)

## Builds on (most cited first, 30 of 36)

- [Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2006.11239/) (2020)
- [High-Resolution Image Synthesis with Latent Diffusion Models](https://mlsanity.com/paper/2112.10752/) (2021)
- [Diffusion Models Beat GANs on Image Synthesis](https://mlsanity.com/paper/2105.05233/) (2021)
- [Segment Anything](https://mlsanity.com/paper/2304.02643/) (2023)
- [Deep Unsupervised Learning using Nonequilibrium Thermodynamics](https://mlsanity.com/paper/1503.03585/) (2015)
- [Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding](https://mlsanity.com/paper/2205.11487/) (2022)
- [Generative Modeling by Estimating Gradients of the Data Distribution](https://mlsanity.com/paper/1907.05600/) (2019)
- [Improved Denoising Diffusion Probabilistic Models](https://mlsanity.com/paper/2102.09672/) (2021)
- [Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection](https://mlsanity.com/paper/2303.05499/) (2023)
- [Video Diffusion Models](https://mlsanity.com/paper/2204.03458/) (2022)
- [Image Super-Resolution via Iterative Refinement](https://mlsanity.com/paper/2104.07636/) (2021)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [DiffWave: A Versatile Diffusion Model for Audio Synthesis](https://mlsanity.com/paper/2009.09761/) (2020)
- [Cascaded Diffusion Models for High Fidelity Image Generation](https://mlsanity.com/paper/2106.15282/) (2021)
- [Improved Techniques for Training Score-Based Generative Models](https://mlsanity.com/paper/2006.09011/) (2020)
- [Grounded SAM: Assembling Open-World Models for Diverse Visual Tasks](https://mlsanity.com/paper/2401.14159/) (2024)
- [Adversarial Diffusion Distillation](https://mlsanity.com/paper/2311.17042/) (2023)
- [Eyes Wide Shut? Exploring the Visual Shortcomings of Multimodal LLMs](https://mlsanity.com/paper/2401.06209/) (2024)
- [Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding](https://mlsanity.com/paper/2311.16922/) (2023)
- [Diffusion Models already have a Semantic Latent Space](https://mlsanity.com/paper/2210.10960/) (2022)
- [Emergent Linear Representations in World Models of Self-Supervised Sequence Models](https://mlsanity.com/paper/2309.00941/) (2023)
- [Training-Free Layout Control with Cross-Attention Guidance](https://mlsanity.com/paper/2304.03373/) (2023)
- [Recognize Anything: A Strong Image Tagging Model](https://mlsanity.com/paper/2306.03514/) (2023)
- [Diffusion Self-Guidance for Controllable Image Generation](https://mlsanity.com/paper/2306.00986/) (2023)
- [FreeDoM: Training-Free Energy-Guided Conditional Diffusion Model](https://mlsanity.com/paper/2303.09833/) (2023)
- [What's "up" with vision-language models? Investigating their struggle with spatial reasoning](https://mlsanity.com/paper/2310.19785/) (2023)
- [What the DAAM: Interpreting Stable Diffusion Using Cross Attention](https://mlsanity.com/paper/2210.04885/) (2022)
- [k-Sparse Autoencoders](https://mlsanity.com/paper/1312.5663/) (2013)
- [Localizing Object-level Shape Variations with Text-to-Image Diffusion Models](https://mlsanity.com/paper/2303.11306/) (2023)
- [Understanding the Latent Space of Diffusion Models through the Lens of Riemannian Geometry](https://mlsanity.com/paper/2307.12868/) (2023)

## Built on by (most cited first, 22 of 22)

- [Probing the Representational Power of Sparse Autoencoders in Vision Models](https://mlsanity.com/paper/2508.11277/) (2025)
- [Temporal Concept Dynamics in Diffusion Models via Prompt-Conditioned Interventions](https://mlsanity.com/paper/2512.08486/) (2025)
- [Revealing the Attention Floating Mechanism in Masked Diffusion Models](https://mlsanity.com/paper/2601.07894/) (2026)
- [Why Do Video Diffusion Models Violate Physics? Unveiling the Flaws in Attention Mechanisms](https://mlsanity.com/paper/2609.23658/) (2026)
- [Focal Guidance: Unlocking Controllability from Semantic-Weak Layers in Video Diffusion Models](https://mlsanity.com/paper/2601.07287/) (2026)
- [Meltdown: Circuits and Bifurcations in Point-Cloud-Conditioned 3D Diffusion Transformers](https://mlsanity.com/paper/2602.11130/) (2026)
- [Embedding-Aware Feature Discovery: Bridging Latent Representations and Interpretable Features in Event Sequences](https://mlsanity.com/paper/2603.15713/) (2026)
- [Measuring Temporal Linguistic Emergence in Diffusion Language Models](https://mlsanity.com/paper/2604.23235/) (2026)
- [Residualized Temporal Sparse Autoencoders for Interpreting Diffusion Models](https://mlsanity.com/paper/2605.27813/) (2026)
- [Self-Attention Decomposition For Training Free Diffusion Editing](https://mlsanity.com/paper/2510.22650/) (2025)
- [FreeFuse: Multi-Subject LoRA Fusion via Adaptive Token-Level Routing at Test Time](https://mlsanity.com/paper/2510.23515/) (2025)
- [Beyond the Black Box: Identifiable Interpretation and Control in Generative Models via Causal Minimality](https://mlsanity.com/paper/2512.10720/) (2025)
- [RAIGen: Rare Attribute Identification in Text-to-Image Generative Models](https://mlsanity.com/paper/2602.06806/) (2026)
- [General and Efficient Steering of Diffusion Models](https://mlsanity.com/paper/2602.11395/) (2026)
- [ATHENA: Adaptive Test-Time Steering for Improving Count Fidelity in Diffusion Models](https://mlsanity.com/paper/2603.19676/) (2026)
- [Robust and Generalizable Safety Steering for Text-to-Image Diffusion Transformers](https://mlsanity.com/paper/2605.30049/) (2026)
- [Information-Theoretic Classifier-Free Guidance with Adaptive Schedule Optimization](https://mlsanity.com/paper/2606.24025/) (2026)
- [Position: Preventing AI-Generated CSAM Necessitates New Approaches to AI Safety](https://mlsanity.com/paper/2607.05407/) (2026)
- [Interpreting Object-Dependent Concept Brittleness in Text-to-Image Diffusion Models](https://mlsanity.com/paper/2609.09909/) (2026)
- [D-Scope: Decomposing and Steering Diffusion Transformers with Sparse Autoencoders](https://mlsanity.com/paper/2609.39625/) (2026)
- [SharedSAE: a unified sparse dictionary across u-net blocks and denoising time for diffusion models](https://mlsanity.com/paper/2610.13997/) (2026)
- [We Have Been Interpreting Diffusion Transformer Backbone Behaviors Wrong](https://mlsanity.com/paper/2610.14517/) (2026)

## Abstract (arXiv)

Diffusion models have become the go-to method for text-to-image generation, producing high-quality images from pure noise. However, the inner workings of diffusion models is still largely a mystery due to their black-box nature and complex, multi-step generation process. Mechanistic interpretability techniques, such as Sparse Autoencoders (SAEs), have been successful in understanding and steering the behavior of large language models at scale. However, the great potential of SAEs has not yet been applied toward gaining insight into the intricate generative process of diffusion models. In this work, we leverage the SAE framework to probe the inner workings of a popular text-to-image diffusion model, and uncover a variety of human-interpretable concepts in its activations. Interestingly, we find that even before the first reverse diffusion step is completed, the final composition of the scene can be predicted surprisingly well by looking at the spatial distribution of activated concepts. Moreover, going beyond correlational analysis, we design intervention techniques aimed at manipulating image composition and style, and demonstrate that (1) in early stages of diffusion image composition can be effectively controlled, (2) in the middle stages image composition is finalized, however stylistic interventions are effective, and (3) in the final stages only minor textural details are subject to change.