mlsanityData to 7 Oct 2026

Common Diffusion Noise Schedules and Sample Steps are Flawed

arXiv 2305.08891 ↗First posted 15 May 2023cs.CVcited by 308

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

We discover that common diffusion noise schedules do not enforce the last timestep to have zero signal-to-noise ratio (SNR), and some implementations of diffusion samplers do not start from the last timestep. Such designs are flawed and do not reflect the fact that the model is given pure Gaussian noise at inference, creating a discrepancy between training and inference. We show that the flawed design causes real problems in existing implementations. In Stable Diffusion, it severely limits the model to only generate images with medium brightness and prevents it from generating very bright and dark samples. We propose a few simple fixes: (1) rescale the noise schedule to enforce zero terminal SNR; (2) train the model with v prediction; (3) change the sampler to always start from the last timestep; (4) rescale classifier-free guidance to prevent over-exposure. These simple changes ensure the diffusion process is congruent between training and inference and allow the model to generate samples more faithful to the original data distribution.

Key facts. Cited by 308 papers in the map (data through 7 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 12 of its 16 references that are in the map. First posted on arXiv 15 May 2023.

Builds on

12 of 16 references.

Closest ancestors · 8

Same direction, newest first

  1. 2022Classifier-Free Diffusion Guidance26 Jul
  2. 2022DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps2 Jun
  3. 2022Pseudo Numerical Methods for Diffusion Models on Manifolds20 Feb
  4. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  5. 2021Improved Denoising Diffusion Probabilistic Models18 Feb
  6. 2020Score-Based Generative Modeling through Stochastic Differential Equations26 Nov
  7. 2020Denoising Diffusion Implicit Models6 Oct
  8. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 4

  1. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  2. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,322
  3. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,131
  4. 2016Improved Techniques for Training GANscited by 2,938

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 42; 2024: 116; 2025: 87; 2026: 63.

308 papers.

Sort:

By area:

  1. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesiscited by 4,290
  2. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesiscited by 4,094
  3. 2023Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasetscited by 2,526
  4. 2024CogVideoX: Text-to-Video Diffusion Models with An Expert Transformercited by 2,198
  5. 2023MVDream: Multi-view Diffusion for 3D Generationcited by 928
  6. 2023Adversarial Diffusion Distillationcited by 741
  7. 2024Lumiere: A Space-Time Diffusion Model for Video Generationcited by 450
  8. 2024CLAY: A Controllable Large-scale Generative Model for Creating High-quality 3D Assetscited by 444
  9. 2025Hunyuan3D 2.0: Scaling Diffusion Models for High Resolution Textured 3D Assets Generationcited by 342
  10. 2023Emu Edit: Precise Image Editing via Recognition and Generation Taskscited by 321
  11. 2023SeeSR: Towards Semantics-Aware Real-World Image Super-Resolutioncited by 311
  12. 2023Photorealistic Video Generation with Diffusion Modelscited by 304
38 more
  1. 2024Open-Sora Plan: Open-Source Large Video Generation Modelcited by 286
  2. 2024GeoWizard: Unleashing the Diffusion Priors for 3D Geometry Estimation from a Single Imagecited by 283
  3. 2024Simplifying, Stabilizing and Scaling Continuous-Time Consistency Modelscited by 276
  4. 2024SDXL-Lightning: Progressive Adversarial Diffusion Distillationcited by 258
  5. 2023ImageDream: Image-Prompt Multi-view Diffusion for 3D Generationcited by 256
  6. 2023Emu Video: Factorizing Text-to-Video Generation by Explicit Image Conditioningcited by 254
  7. 2023Pixel-Aware Stable Diffusion for Realistic Image Super-resolution and Personalized Stylizationcited by 251
  8. 2024Playground v2.5: Three Insights towards Enhancing Aesthetic Quality in Text-to-Image Generationcited by 249
  9. 2024NaturalSpeech 3: Zero-Shot Speech Synthesis with Factorized Codec and Diffusion Modelscited by 222
  10. 2024CRM: Single Image to 3D Textured Mesh with Convolutional Reconstruction Modelcited by 214
  11. 2024Motion-I2V: Consistent and Controllable Image-to-Video Generation with Explicit Motion Modelingcited by 209
  12. 2024How Far is Video Generation from World Model: A Physical Law Perspectivecited by 204
  13. 2024Fast Timing-Conditioned Latent Audio Diffusioncited by 178
  14. 2025History-Guided Video Diffusioncited by 165
  15. 2024RGBleftrightarrowX: Image decomposition and synthesis using material- and lighting-aware diffusion modelscited by 159
  16. 2024MaskGCT: Zero-Shot Text-to-Speech with Masked Generative Codec Transformercited by 159
  17. 2025Hunyuan3D 2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Materialcited by 157
  18. 2025Diffusion Adversarial Post-Training for One-Step Video Generationcited by 142
  19. 2024Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Imagecited by 132
  20. 2024ConsistI2V: Enhancing Visual Consistency for Image-to-Video Generationcited by 130
  21. 2023MagicPose: Realistic Human Poses and Facial Expressions Retargeting with Identity-aware Diffusioncited by 115
  22. 2024Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Modelscited by 107
  23. 2024MindEye2: Shared-Subject Models Enable fMRI-To-Image With 1 Hour of Datacited by 105
  24. 2025One-Minute Video Generation with Test-Time Trainingcited by 100
  25. 2023Guided Flows for Generative Modeling and Decision Makingcited by 99
  26. 2024MVDiffusion++: A Dense High-resolution Multi-view Diffusion Model for Single or Sparse-view 3D Object Reconstructioncited by 94
  27. 2023FreeInit: Bridging Initialization Gap in Video Diffusion Modelscited by 93
  28. 2023Stay on topic with Classifier-Free Guidancecited by 90
  29. 2024V3D: Video Diffusion Models are Effective 3D Generatorscited by 88
  30. 2024Long-form music generation with latent diffusioncited by 88
  31. 2024Large Concept Models: Language Modeling in a Sentence Representation Spacecited by 86
  32. 2024Arbitrary-steps Image Super-resolution via Diffusion Inversioncited by 85
  33. 2025Autoregressive Adversarial Post-Training for Real-Time Interactive Video Generationcited by 84
  34. 2024Fine-Tuning Image-Conditional Diffusion Models is Easier than You Thinkcited by 82
  35. 2023DreamTime: An Improved Optimization Strategy for Diffusion-Guided 3D Generationcited by 81
  36. 2023FlowVid: Taming Imperfect Optical Flows for Consistent Video-to-Video Synthesiscited by 79
  37. 2023Training-free Linear Image Inverses via Flowscited by 70
  38. 2023HyperHuman: Hyper-Realistic Human Generation with Latent Structural Diffusioncited by 68

50 most cited of 308 All 308 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Constant Rate Scheduling: A General Framework for Optimizing Diffusion Noise Schedule via Distributional Change
  2. 2025It Takes Two to Tango: Two Parallel Samplers Improve Quality in Diffusion Models for Limited Steps
  3. 2023Unmasking Bias in Diffusion Model Training
  4. 2024Not All Noises Are Created Equally:Diffusion Noise Selection and Optimization
  5. 2025A Comprehensive Review on Noise Control of Diffusion Model
  6. 2024Consistent Diffusion Meets Tweedie: Training Exact Ambient Diffusion Models with Noisy Data
  7. 2025Is Your Conditional Diffusion Model Actually Denoising?
  8. 2026Couple to Control: Joint Initial Noise Design in Diffusion Models