mlsanityData to 7 Oct 2026

What the DAAM: Interpreting Stable Diffusion Using Cross Attention

arXiv 2210.04885 ↗First posted 10 Oct 2022cs.CV cs.CLcited by 233

On the map: Diffusion models → Image generation, text-to-image and editing

Abstract (arXiv)

Large-scale diffusion neural networks represent a substantial milestone in text-to-image generation, but they remain poorly understood, lacking interpretability analyses. In this paper, we perform a text-image attribution analysis on Stable Diffusion, a recently open-sourced model. To produce pixel-level attribution maps, we upscale and aggregate cross-attention word-pixel scores in the denoising subnetwork, naming our method DAAM. We evaluate its correctness by testing its semantic segmentation ability on nouns, as well as its generalized attribution quality on all parts of speech, rated by humans. We then apply DAAM to study the role of syntax in the pixel space, characterizing head--dependent heat map interaction patterns for ten common dependency relations. Finally, we study several semantic phenomena using DAAM, with a focus on feature entanglement, where we find that cohyponyms worsen generation quality and descriptive adjectives attend too broadly. To our knowledge, we are the first to interpret large diffusion models from a visuolinguistic perspective, which enables future lines of research. Our code is at https://github.com/castorini/daam.

Key facts. Cited by 233 papers in the map (data through 7 Oct 2026). Placed in Image generation, text-to-image and editing, one of 14 directions in Diffusion models. Builds on 36 of its 61 references that are in the map. First posted on arXiv 10 Oct 2022.

Builds on

36 of 61 references.

Closest ancestors · 2

Same direction, newest first

  1. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding23 May
  2. 2022Hierarchical Text-Conditional Image Generation with CLIP Latents13 Apr

Other references · 34

  1. 2017Attention Is All You Needcited by 59,932
  2. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  3. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  4. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  5. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,604
  6. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  7. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  8. 2021Swin Transformer: Hierarchical Vision Transformer using Shifted Windowscited by 12,082
  9. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 8,964
  10. 2017Mask R-CNNcited by 7,259
  11. 2016Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localizationcited by 6,690
  12. 2018A Style-Based Generator Architecture for Generative Adversarial Networkscited by 5,986
22 more
  1. 2016Feature Pyramid Networks for Object Detectioncited by 5,021
  2. 2013Auto-Encoding Variational Bayescited by 3,980
  3. 2022Prompt-to-Prompt Image Editing with Cross Attention Controlcited by 2,409
  4. 2021Masked-attention Mask Transformer for Universal Image Segmentationcited by 2,338
  5. 2022DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Stepscited by 1,961
  6. 2019MMDetection: Open MMLab Detection Toolbox and Benchmarkcited by 1,379
  7. 2022Diffusion Models: A Comprehensive Survey of Methods and Applicationscited by 1,107
  8. 2019What Does BERT Look At? An Analysis of BERT's Attentioncited by 994
  9. 2022Pseudo Numerical Methods for Diffusion Models on Manifoldscited by 695
  10. 2021Image Segmentation Using Text and Image Promptscited by 509
  11. 2019Is Attention Interpretable?cited by 387
  12. 2018On the Robustness of Interpretability Methodscited by 310
  13. 2019Revealing the Dark Secrets of BERTcited by 300
  14. 2022Unsupervised Semantic Segmentation by Distilling Feature Correspondencescited by 241
  15. 2021PiCIE: Unsupervised Semantic Segmentation using Invariance and Equivariance in Clusteringcited by 163
  16. 2021Instances as Queriescited by 157
  17. 2020PhraseCut: Language-based Image Segmentation in the Wildcited by 116
  18. 2021Probing Image-Language Transformers for Verb Understandingcited by 109
  19. 2017Open Vocabulary Scene Parsingcited by 76
  20. 2019AllenNLP Interpret: A Framework for Explaining Predictions of NLP Modelscited by 74
  21. 2022Disentangling visual and written concepts in CLIPcited by 57
  22. 2016Neural Architectures for Fine-grained Entity Type Classificationcited by 15

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 1; 2023: 38; 2024: 69; 2025: 61; 2026: 64.

233 papers.

Sort:

By area:

  1. 2023MasaCtrl: Tuning-Free Mutual Self-Attention Control for Consistent Image Synthesis and Editingcited by 722
  2. 2024MS-Diffusion: Multi-subject Zero-shot Image Personalization with Layout Guidancecited by 142
  3. 2024Lumina-T2X: Transforming Text into Any Modality, Resolution, and Duration via Flow-based Large Diffusion Transformerscited by 137
  4. 2023Editing Implicit Assumptions in Text-to-Image Diffusion Modelscited by 128
  5. 2023Receler: Reliable Concept Erasing of Text-to-Image Diffusion Models via Lightweight Eraserscited by 118
  6. 2023Dataset Diffusion: Diffusion-based Synthetic Dataset Generation for Pixel-Level Semantic Segmentationcited by 112
  7. 2024MindEye2: Shared-Subject Models Enable fMRI-To-Image With 1 Hour of Datacited by 105
  8. 2023Text-to-Image Diffusion Models can be Easily Backdoored through Multimodal Data Poisoningcited by 91
  9. 2023Expressive Text-to-Image Generation with Rich Textcited by 89
  10. 2024ZigMa: A DiT-style Zigzag Mamba Diffusion Modelcited by 84
  11. 2023Open-vocabulary Object Segmentation with Diffusion Modelscited by 82
  12. 2023Divide & Bind Your Attention for Improved Generative Semantic Nursingcited by 80
38 more
  1. 2025Your Large Vision-Language Model Only Needs A Few Attention Heads For Visual Groundingcited by 78
  2. 2023Diffusion Model is Secretly a Training-free Open Vocabulary Semantic Segmentercited by 76
  3. 2025SAeUron: Interpretable Concept Unlearning in Diffusion Models with Sparse Autoencoderscited by 75
  4. 2023Directed Diffusion: Direct Control of Object Placement through Attention Guidancecited by 69
  5. 2023Diffusion Models for Open-Vocabulary Segmentationcited by 63
  6. 2023Focus on Your Instruction: Fine-grained and Multi-instruction Image Editing by Attention Modulationcited by 61
  7. 2024PromptCharm: Text-to-Image Generation through Multi-modal Prompting and Refinementcited by 61
  8. 2024Unveiling and Mitigating Memorization in Text-to-image Diffusion Models through Cross Attentioncited by 57
  9. 2023CONFORM: Contrast is All You Need For High-Fidelity Text-to-Image Diffusion Modelscited by 54
  10. 2024Auffusion: Leveraging the Power of Diffusion and Large Language Models for Text-to-Audio Generationcited by 54
  11. 2024Explainable and Interpretable Multimodal Large Language Models: A Comprehensive Surveycited by 52
  12. 2024Understanding the Impact of Negative Prompts: When and How Do They Take Effect?cited by 50
  13. 2024Towards Understanding the Working Mechanism of Text-to-Image Diffusion Modelcited by 48
  14. 2025ConceptAttention: Diffusion Transformers Learn Highly Interpretable Featurescited by 48
  15. 2024Training-Free Open-Vocabulary Segmentation with Offline Diffusion-Augmented Prototype Generationcited by 43
  16. 2024EmerDiff: Emerging Pixel-level Semantic Knowledge in Diffusion Modelscited by 42
  17. 2023TokenCompose: Text-to-Image Diffusion with Token-level Supervisioncited by 40
  18. 2024Membership Inference on Text-to-Image Diffusion Models via Conditional Likelihood Discrepancycited by 38
  19. 2023Farm3D: Learning Articulated 3D Animals by Distilling 2D Diffusioncited by 37
  20. 2023Interpretable Diffusion via Information Decompositioncited by 37
  21. 2023Geometry-Complete Diffusion for 3D Molecule Generation and Optimizationcited by 36
  22. 2023Unsupervised Compositional Concepts Discovery with Text-to-Image Generative Modelscited by 36
  23. 2024Image Segmentation in Foundation Model Era: A Surveycited by 35
  24. 2025Visual Large Language Models for Generalized and Specialized Applicationscited by 35
  25. 2025A Survey on Mechanistic Interpretability for Multi-Modal Foundation Modelscited by 34
  26. 2024Learning Action and Reasoning-Centric Image Editing from Videos and Simulationscited by 32
  27. 2024Attention Calibration for Disentangled Text-to-Image Personalizationcited by 31
  28. 2024FaithDiff: Unleashing Diffusion Priors for Faithful Image Super-resolutioncited by 31
  29. 2024LLM4GEN: Leveraging Semantic Representation of LLMs for Text-to-Image Generationcited by 30
  30. 2024Open-Vocabulary Attention Maps with Token Optimization for Semantic Segmentation in Diffusion Modelscited by 29
  31. 2024IllusionVQA: A Challenging Optical Illusion Dataset for Vision Language Modelscited by 29
  32. 2024Diffusion Lens: Interpreting Text Encoders in Text-to-Image Pipelinescited by 28
  33. 2024Open-World Human-Object Interaction Detection via Multi-modal Promptscited by 28
  34. 2023Stable Diffusion is Unstablecited by 27
  35. 2024On Mechanistic Knowledge Localization in Text-to-Image Generative Modelscited by 27
  36. 2024MC2: Multi-concept Guidance for Customized Multi-concept Generationcited by 26
  37. 2024DiffCut: Catalyzing Zero-Shot Semantic Segmentation with Diffusion Features and Recursive Normalized Cutcited by 26
  38. 2023Stable Diffusion Exposed: Gender Bias from Prompt to Imagecited by 25

50 most cited of 233 All 233 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Enhancing Semantic Fidelity in Text-to-Image Synthesis: Attention Regulation in Diffusion Models
  2. 2023Are Diffusion Models Vision-And-Language Reasoners?
  3. 2026Omni-Diffusion: Unified Multimodal Understanding and Generation with Masked Discrete Diffusion
  4. 2026Memorization In Stable Diffusion Is Unexpectedly Driven by CLIP Embeddings
  5. 2023The Hidden Language of Diffusion Models
  6. 2026Temporally-Resolved Token Attribution Reveals the Generation Dynamics of Diffusion Language Models
  7. 2024Decoding Diffusion: A Scalable Framework for Unsupervised Analysis of Latent Space Biases and Representations Using Natural Language Prompts
  8. 2024Attention Overlap Is Responsible for The Entity Missing Problem in Text-to-image Diffusion Models!