🌀 Awesome Generative Explainability

This repository contains frontier research on explainable generative models, with a strong emphasis on diffusion-based models. As generative AI continues to evolve, understanding how and why these models generate outputs is critical for building transparent, safe, and controllable AI systems.

---

2026

DifFRACT: Diffusion Feature Reconstruction and Attribution for Circuit Tracing, 2026
[Paper]

Image Generation from Contextually-Contradictory Prompts, CVPR 2026
[Paper]

Diagnosing and Correcting Concept Omission in Multimodal Diffusion Transformers, ICML 2026
[Paper]

When Do Diffusion Models Learn to Generate Multiple Objects?, ICML 2026
[Paper]

Enhancing MMDiT-Based Text-to-Image Models for Similar Subject Generation, TPAMI 2026
[Paper] [Code]

Attention, May I Have Your Decision? Localizing Generative Choices in Diffusion Models, CVPR 2026
[Paper] [Code]

Self-Corrected Image Generation with Explainable Latent Rewards, CVPR 2026
[Paper]

RAISE: Requirement-Adaptive Evolutionary Refinement for Training-Free Text-to-Image Alignment, CVPR 2026
[Paper] [Code]

Erasing Thousands of Concepts: Towards Scalable and Practical Concept Erasure for Text-to-Image Diffusion Models, CVPR 2026
[Paper]

Memorization In Stable Diffusion Is Unexpectedly Driven by CLIP Embeddings, CVPR 2026 Findings
[Paper]

GeoDiv: Framework for Measuring Geographical Diversity in Text-to-Image Models, ICLR 2026
[Paper] [Code] [Project]

Temporal Concept Dynamics in Diffusion Models via Prompt-Conditioned Interventions, ICLR 2026
[Paper] [Code] [Project]

Concept-TRAK: Understanding how Diffusion Models Learn Concepts through Concept-Level Attribution, ICLR 2026
[Paper] [Code]

RAIGen: Rare Attribute Identification in Text-to-Image Generative Models, ICML 2026
[Paper]

SAEmnesia: Erasing Concepts in Diffusion Models with Supervised Sparse Autoencoders, ICML 2026
[Paper] [Code]

DLM-Scope: Mechanistic Interpretability of Diffusion Language Models via Sparse Autoencoders, ICML 2026
[Paper]

SHIFT: Steering Hidden Intermediates in Flow Transformers, ICML 2026 Workshop on Mechanistic Interpretability
[Paper]

DreamReader: An Interpretability Toolkit for Text-to-Image Models, 2026
[Paper]

Diff-Aid: Inference-time Adaptive Interaction Denoising for Rectified Text-to-Image Generation, 2026
[Paper]

ELROND: Exploring and Decomposing Intrinsic Capabilities of Diffusion Models, 2026
[Paper]

Unraveling MMDiT Blocks: Training-free Analysis and Enhancement of Text-conditioned Diffusion, 2026
[Paper]

Leveraging Semantic Attribute Binding for Free-Lunch Color Control in Diffusion Models, WACV 2026
[Paper] [Project]

2025

Seg4Diff: Unveiling Open-Vocabulary Segmentation in Text-to-Image Diffusion Transformers, NeurIPS 2025
[Paper] [Code] [Project]

Revelio: Interpreting and Leveraging Semantic Information in Diffusion Models, ICCV 2025
[Paper] [Code]

Localizing Knowledge in Diffusion Transformers, NeurIPS 2025
[Paper] [Code] [Project]

Precise Parameter Localization for Textual Generation in Diffusion Models, ICLR 2025
[Paper] [Project]

Dissecting and Mitigating Diffusion Bias via Mechanistic Interpretability, CVPR 2025
[Paper] [Code]

Text Embedding is Not All You Need: Attention Control for Text-to-Image Semantic Alignment with Text Self-Attention Maps, CVPR 2025
[Paper]

Progressive Compositionality in Text-to-Image Generative Models, ICLR 2025
[Paper]

Multi-Class Textual-Inversion Secretly Yields a Semantic-Agnostic Classifier, WACV 2025
[Paper] [Code]

One-Prompt-One-Story: Free-Lunch Consistent Text-to-Image Generation Using a Single Prompt, ICLR 2025
[Paper] [Code] [Project]

Emergent Temporal Correspondences from Video Diffusion Transformers, NeurIPS 2025
[Paper] [Code] [Project]

Make It Count: Text-to-Image Generation with an Accurate Number of Objects, CVPR 2025
[Paper] [Code] [Project]

Cross-Attention Head Position Patterns Can Align with Human Visual Concepts in Text-to-Image Generative Models, ICLR 2025
[Paper] [Code]

Decoding Vision Transformers: the Diffusion Steering Lens, CVPR 2025 Workshop on Mechanistic Interpretability for Vision
[Paper]

Add-it: Training-Free Object Insertion in Images with Pretrained Diffusion Models, ICLR 2025
[Paper] [Code] [Project]

Emergence and Evolution of Interpretable Concepts in Diffusion Models, NeurIPS 2025
[Paper] [Code]

Interpretable Generative Models through Post-hoc Concept Bottlenecks, CVPR 2025
[Paper]

Show and Tell: Visually Explainable Deep Neural Nets via Spatially-Aware Concept Bottleneck Models, CVPR 2025
[Paper] [Project]

Controlling Language and Diffusion Models by Transporting Activations, ICLR 2025
[Paper] [Code]

Restyling Unsupervised Concept Based Interpretable Networks with Generative Models, ICLR 2025 [Paper] [Project]

A General Framework for Inference-time Scaling and Steering of Diffusion Models, ICML 2025 [Paper] [Code]

ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features, ICML 2025 [Paper] [Code] 2024 venue-aware update

ABE-CLIP: Training-Free Attribute Binding Enhancement for Compositional Image-Text Matching, 2025

What Drives Compositional Generalization? The Importance of Continuous Training Objectives in Visual Generative Models, 2025

Representation Entanglement for Generation: Training Diffusion Transformers Is Much Easier Than You Think, 2025

Adapting Self-Supervised Representations as a Latent Space for Efficient Generation, 2025

LumiCtrl: Learning Illuminant Prompts for Lighting Control in Personalized Text-to-Image Models, 2025

Understanding Attention Mechanism in Video Diffusion Models, 2025

DeLeaker: Dynamic Inference-Time Reweighting for Semantic Leakage Mitigation in Text-to-Image Models, 2025

RefAM: Attention Magnets for Zero-Shot Referral Segmentation, 2025

No Other Representation Component Is Needed: Diffusion Transformers Can Provide Representation Guidance by Themselves
[Paper] [Code]

SAeUron: Interpretable Concept Unlearning in Diffusion Models with Sparse Autoencoders
[Paper] [Code]

TIDE: Temporal-Aware Sparse Autoencoders for Interpretable Diffusion Transformers in Image Generation
[Paper]

Concept Steerers: Leveraging K-Sparse Autoencoders for Controllable Generations
[Paper] [Code]

Sparse Autoencoder as a Zero-Shot Classifier for Concept Erasing in Text-to-Image Diffusion Models
[Paper] [Code]

2024

On Mechanistic Knowledge Localization in Text-to-Image Generative Models, ICML 2024
[Paper] [Project]

Localizing and Editing Knowledge in Text-to-Image Generative Models, ICLR 2024
[Paper]

Interpretable Diffusion via Information Decomposition, ICLR 2024
[Paper] [Code]

Get What You Want, Not What You Don't: Image Content Suppression for Text-to-Image Diffusion Models, ICLR 2024
[Paper] [Code]

Token Merging for Training-Free Semantic Binding in Text-to-Image Synthesis, NeurIPS 2024
[Paper] [Code]

Faster Diffusion: Rethinking the Role of UNet Encoder in Diffusion Models, NeurIPS 2024
[Paper]

Towards Understanding the Working Mechanism of Text-to-Image Diffusion Model, NeurIPS 2024
[Paper]

Diffusion Lens: Interpreting Text Encoders in Text-to-Image Pipelines, ACL 2024
[Paper] [Code]

Discovering Interpretable Directions in the Semantic Latent Space of Diffusion Models, FG 2024
[Paper]

Training-Free Layout Control with Cross-Attention Guidance, WACV 2024
[Paper] [Code]

Understanding Hallucinations in Diffusion Models through Mode Interpolation, NeurIPS 2024
[Paper]

Unified Concept Editing in Diffusion Models, WACV 2024
[Paper] [Project]

Reliable and Efficient Concept Erasure of Text-to-Image Diffusion Models, ECCV 2024
[Paper] [Code]

Exploring Diffusion Time-steps for Unsupervised Representation Learning, ICLR 2024
[Paper] [Code]

Explainable Generative AI (GenXAI): A Survey, Conceptualization, and Research Agenda
[Paper]

Generated Bias: Auditing Racial Bias in Text-to-Image Diffusion Models, ECCV 2024 Workshop
[Paper]

Concept Bottleneck Generative Models, ICLR 2024
[Paper]

Generalization in Diffusion Models Arises from Geometry-Adaptive Harmonic Representations, ICLR 2024 Oral
[Paper] [Code]

ConceptExpress: Harnessing Diffusion Models for Single-image Unsupervised Concept Extraction, ECCV 2024 Oral
[Paper] [Code]

Self-Discovering Interpretable Diffusion Latent Directions for Responsible Text-to-Image Generation, CVPR 2024
[Paper] [Code]

PreciseControl: Enhancing Text-To-Image Diffusion Models with Fine-Grained Attribute Control, ECCV 2024
[Paper] [Project]

Concept Sliders: LoRA Adaptors for Precise Control in Diffusion Models, ECCV 2024
[Paper] [Code]

Diffexplainer: Towards Cross-modal Global Explanations with Diffusion Models
[Paper]

Explaining Generative Diffusion Models via Visual Analysis for Interpretable Decision-making Process
[Paper]

STEREO: A Two-Stage Framework for Adversarially Robust Concept Erasing from Text-to-Image Diffusion Models, CVPR 2025 Highlight
[Paper] [Code]

Trade-offs in Fine-tuned Diffusion Models Between Accuracy and Interpretability
[Paper]

Unpacking SDXL Turbo: Interpreting Text-to-Image Models with Sparse Autoencoders
[Paper] [Code]

Diffusion Models Learn Low-Dimensional Distributions via Subspace Clustering
[Paper] [Code]

Exploring low-dimensional subspaces in diffusion models for controllable image editing
[Paper] [Code]

Interpreting the Weight Space of Customized Diffusion Models, NeurIPS 2024
[Paper] [Project]

2023

A Tale of Two Features: Stable Diffusion Complements DINO for Zero-Shot Semantic Correspondence, NeurIPS 2023

Diffusion Self-Guidance for Controllable Image Generation, NeurIPS 2023
[Paper] [Code]

Null-text Inversion for Editing Real Images using Guided Diffusion Models, CVPR 2023
[Paper] [Code]

Prompt-to-Prompt Image Editing with Cross Attention Control, ACL 2023
[Paper] [Code]

Attend-and-Excite: Attention-Based Semantic Guidance for Text-to-Image Diffusion Models, SIGGRAPH 2023
[Paper] [Code]

Concept Algebra for (Score-Based) Text-Controlled Generative Models, NeurIPS 2023
[Paper] [Code]

Cones: Concept Neurons in Diffusion Models for Customized Generation, ICML 2023 Oral
[Paper]

Unsupervised Compositional Concepts Discovery with Text-to-Image Generative Models, ICCV 2023
[Paper]

Your Diffusion Model is Secretly a Zero-Shot Classifier
[Paper] [Code]

Diffusion models already have a semantic latent space, ICLR 2023
[Paper] [Code]

Understanding the latent space of diffusion models through the lens of riemannian geometry, NeurIPS 2023
[Paper] [Code]

Emergent Correspondence from Image Diffusion, NeurIPS 2023
[Paper] [Code]

2022 and Earlier

An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion
[Paper] [Code]

What the DAAM: Interpreting Stable Diffusion Using Cross Attention
[Paper] [Code]

CLIPDraw: Exploring Text-to-Drawing Synthesis through Language-Image Encoders, NeurIPS 2022
[Paper] [Code]

Understanding Diffusion Models: A Unified Perspective, 2022
[Paper]

Discovering Latent Concepts Learned in BERT, ICLR 2022
[Paper] [Code]

GAN Dissection: Visualizing and Understanding Generative Adversarial Networks, ICLR 2019
[Paper] [Code]

GANSpace: Discovering Interpretable GAN Controls, NeurIPS 2020
[Paper] [Code]