mlsanityData to 14 Oct 2026

Exploiting the Signal-Leak Bias in Diffusion Models

arXiv 2309.15842 ↗First posted 27 Sep 2023cs.CV cs.LGcited by 36

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

There is a bias in the inference pipeline of most diffusion models. This bias arises from a signal leak whose distribution deviates from the noise distribution, creating a discrepancy between training and inference processes. We demonstrate that this signal-leak bias is particularly significant when models are tuned to a specific style, causing sub-optimal style matching. Recent research tries to avoid the signal leakage during training. We instead show how we can exploit this signal-leak bias in existing diffusion models to allow more control over the generated images. This enables us to generate images with more varied brightness, and images that better match a desired style or color. By modeling the distribution of the signal leak in the spatial frequency and pixel domains, and including a signal leak in the initial latent, we generate images that better match expected results without any additional training.

Key facts. Cited by 36 papers in the map (data through 14 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 24 of its 45 references that are in the map. First posted on arXiv 27 Sep 2023.

Builds on

24 of 45 references.

Closest ancestors · 9

Same direction, newest first

  1. 2023Alleviating Exposure Bias in Diffusion Models through Sampling with Shifted Time Steps24 May
  2. 2023Common Diffusion Noise Schedules and Sample Steps are Flawed15 May
  3. 2023Input Perturbation Reduces Exposure Bias in Diffusion Models27 Jan
  4. 2022Classifier-Free Diffusion Guidance26 Jul
  5. 2022Pseudo Numerical Methods for Diffusion Models on Manifolds20 Feb
  6. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  7. 2021Improved Denoising Diffusion Probabilistic Models18 Feb
  8. 2020Denoising Diffusion Implicit Models6 Oct
  9. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 15

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  3. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,355
  4. 2015Rethinking the Inception Architecture for Computer Visioncited by 5,475
  5. 2021GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Modelscited by 3,614
  6. 2022DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generationcited by 3,300
  7. 2022An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversioncited by 2,245
  8. 2021SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equationscited by 1,937
  9. 2015Sequence Level Training with Recurrent Neural Networkscited by 576
  10. 2022GeoDiff: a Geometric Diffusion Model for Molecular Conformation Generationcited by 516
  11. 2019Generalization in Generation: A closer look at Exposure Biascited by 123
  12. 2023LDM3D: Latent Diffusion Model for 3Dcited by 53
3 more
  1. 2022Shifted Diffusion for Text-to-image Generationcited by 37
  2. 2020Stochastic Frequency Masking to Improve Super-Resolution and Denoising Networkscited by 21
  3. 2022Arbitrary Style Guidance for Enhanced Diffusion-Based Text-to-Image Generationcited by 9

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 3; 2024: 12; 2025: 15; 2026: 6.

36 papers.

Sort:

By area:

  1. 2024StableNormal: Reducing Diffusion Variance for Stable and Sharp Normalcited by 154
  2. 2025Hi3DGen: High-fidelity 3D Geometry Generation from Images via Normal Bridgingcited by 99
  3. 2023FreeInit: Bridging Initialization Gap in Video Diffusion Modelscited by 95
  4. 2024You See it, You Got it: Learning 3D Creation on Pose-Free Videos at Scalecited by 65
  5. 2024All Seeds Are Not Equal: Enhancing Compositional Text-to-Image Generation with Reliable Random Seedscited by 31
  6. 2024Diffusion Posterior Proximal Sampling for Image Restorationcited by 14
  7. 2024I4VGen: Image as Free Stepping Stone for Text-to-Video Generationcited by 14
  8. 2023DREAM: Diffusion Rectification and Estimation-Adaptive Modelscited by 12
  9. 2025FDS: Frequency-Aware Denoising Score for Text-Guided Latent Diffusion Image Editingcited by 12
  10. 2025Free-Lunch Color-Texture Disentanglement for Stylized Image Generationcited by 8
  11. 2024SCP-Diff: Spatial-Categorical Joint Prior for Diffusion Based Semantic Image Synthesiscited by 7
  12. 2023One More Step: A Versatile Plug-and-Play Module for Rectifying Diffusion Schedule Flaws and Enhancing Low-Frequency Controlscited by 6
24 more
  1. 2024FLAME Diffuser: Wildfire Image Synthesis using Mask Guided Diffusioncited by 5
  2. 2025Enhancing Diffusion Model Stability for Image Restoration via Gradient Managementcited by 5
  3. 2024MRPD: Undersampled MRI reconstruction by prompting a large latent diffusion modelcited by 4
  4. 2025Antithetic Noise in Diffusion Modelscited by 4
  5. 2026Exposure Bias Can Alleviate Itself via Directional and Frequency Rectification in Flow Matchingcited by 4
  6. 2024Latent Space Disentanglement in Diffusion Transformers Enables Precise Zero-shot Semantic Editingcited by 3
  7. 2024Mitigating Data Consistency Induced Discrepancy in Cascaded Diffusion Models for Sparse-view CT Reconstructioncited by 2
  8. 2024Solving Prior Distribution Mismatch in Diffusion Models via Optimal Transportcited by 2
  9. 2025SAGA: Learning Signal-Aligned Distributions for Improved Text-to-Image Generationcited by 2
  10. 2025Diffusion-Driven Progressive Target Manipulation for Source-Free Domain Adaptationcited by 2
  11. 2026Training-free image inversion for one-step diffusion modelscited by 2
  12. 2024Latent Space Disentanglement in Diffusion Transformers Enables Zero-shot Fine-grained Semantic Editingcited by 1
  13. 2025Diffusion Prism: Enhancing Diversity and Morphology Consistency in Mask-to-Image Diffusioncited by 1
  14. 2025Text-to-Image Alignment in Denoising-Based Models through Step Selectioncited by 1
  15. 2025Ctrl-Z Sampling: Scaling Diffusion Sampling with Controlled Random Zigzag Explorationscited by 1
  16. 2025LoRAShield: Data-Free Editing Alignment for Secure Personalized LoRA Sharingcited by 1
  17. 2025Example-Based Feature Painting on Texturescited by 1
  18. 2026Bias mitigation in graph diffusion modelscited by 1
  19. 2025DidSee: Diffusion-Based Depth Completion for Material-Agnostic Robotic Perception and Manipulationcited by 0
  20. 2025Proto-LeakNet: Towards Signal-Leak Aware Attribution in Synthetic Human Face Imagerycited by 0
  21. 2025ReflexFlow: Rethinking Learning Objective for Exposure Bias Alleviation in Flow Matchingcited by 0
  22. 2026FreqTrack: Frequency Learning based Vision Transformer for RGB-Event Object Trackingcited by 0
  23. 2026Mechanisms of Misgeneralization in Physical Sequence Modelingcited by 0
  24. 2026Anchoring and Steering Diffusion: Enhancing the Faithfulness of Text-to-Image Generation at Inference Timecited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2023Unmasking Bias in Diffusion Model Training
  2. 2026Diffusion Models, Denoiser Architecture and Creativity
  3. 2025Deeper Diffusion Models Amplify Bias
  4. 2024Suppress Content Shift: Better Diffusion Features via Off-the-Shelf Generation Techniques
  5. 2023Diffusion Model with Perceptual Loss
  6. 2023Masked Diffusion Models Are Fast Distribution Learners
  7. 2025Shaping Inductive Bias in Diffusion Models through Frequency-Based Noise Control
  8. 2024Simplified and Generalized Masked Diffusion for Discrete Data