mlsanityData to 14 Oct 2026

Improved Distribution Matching Distillation for Fast Image Synthesis

arXiv 2405.14867 ↗First posted 23 May 2024cs.CVcited by 684

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

Recent approaches have shown promises distilling diffusion models into efficient one-step generators. Among them, Distribution Matching Distillation (DMD) produces one-step generators that match their teacher in distribution, without enforcing a one-to-one correspondence with the sampling trajectories of their teachers. However, to ensure stable training, DMD requires an additional regression loss computed using a large set of noise-image pairs generated by the teacher with many steps of a deterministic sampler. This is costly for large-scale text-to-image synthesis and limits the student's quality, tying it too closely to the teacher's original sampling paths. We introduce DMD2, a set of techniques that lift this limitation and improve DMD training. First, we eliminate the regression loss and the need for expensive dataset construction. We show that the resulting instability is due to the fake critic not estimating the distribution of generated samples accurately and propose a two time-scale update rule as a remedy. Second, we integrate a GAN loss into the distillation procedure, discriminating between generated samples and real images. This lets us train the student model on real data, mitigating the imperfect real score estimation from the teacher model, and enhancing quality. Lastly, we modify the training procedure to enable multi-step sampling. We identify and address the training-inference input mismatch problem in this setting, by simulating inference-time generator samples during training time. Taken together, our improvements set new benchmarks in one-step image generation, with FID scores of 1.28 on ImageNet-64x64 and 8.35 on zero-shot COCO 2014, surpassing the original teacher despite a 500X reduction in inference cost. Further, we show our approach can generate megapixel images by distilling SDXL, demonstrating exceptional visual quality among few-step methods.

Key facts. Cited by 684 papers in the map (data through 14 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 81 of its 91 references that are in the map. First posted on arXiv 23 May 2024.

Builds on

81 of 91 references.

Closest ancestors · 36

Same direction, newest first

  1. 2024Imagine Flash: Accelerating Emu Diffusion Models with Backward Distillation8 May
  2. 2024Accelerating Image Generation with Sub-path Linear Approximation Model22 Apr
  3. 2024Hyper-SD: Trajectory Segmented Consistency Model for Efficient Image Synthesis21 Apr
  4. 2024Score identity Distillation: Exponentially Fast Distillation of Pretrained Diffusion Models for One-Step Generation5 Apr
  5. 2024Fast High-Resolution Image Synthesis with Latent Adversarial Diffusion Distillation18 Mar
  6. 2024Multistep Consistency Models11 Mar
  7. 2024Trajectory Consistency Distillation: Improved Latent Consistency Distillation by Semi-Linear Consistency Function with Trajectory Mapping29 Feb
  8. 2024SDXL-Lightning: Progressive Adversarial Diffusion Distillation21 Feb
  9. 2023One-step Diffusion with Distribution Matching Distillation30 Nov
  10. 2023Adversarial Diffusion Distillation28 Nov
  11. 2023LCM-LoRA: A Universal Stable-Diffusion Acceleration Module9 Nov
  12. 2023Improved Techniques for Training Consistency Models22 Oct
24 more
  1. 2023Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference6 Oct
  2. 2023Consistency Trajectory Models: Learning Probability Flow ODE Trajectory of Diffusion1 Oct
  3. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis4 Jul
  4. 2023BOOT: Data-free Distillation of Denoising Diffusion Models with Bootstrapping8 Jun
  5. 2023Diff-Instruct: A Universal Approach for Transferring Knowledge From Pre-trained Diffusion Models29 May
  6. 2023Unifying GANs and Score-Based Diffusion as Generative Particle Models25 May
  7. 2023The Score-Difference Flow for Implicit Generative Modeling25 Apr
  8. 2023TRACT: Denoising Diffusion Models with Transitive Closure Time-Distillation7 Mar
  9. 2023Consistency Models2 Mar
  10. 2023UniPC: A Unified Predictor-Corrector Framework for Fast Sampling of Diffusion Models9 Feb
  11. 2022Fast Sampling of Diffusion Models via Operator Learning24 Nov
  12. 2022DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models2 Nov
  13. 2022On Distillation of Guided Diffusion Models6 Oct
  14. 2022Classifier-Free Diffusion Guidance26 Jul
  15. 2022DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps2 Jun
  16. 2022Elucidating the Design Space of Diffusion-Based Generative Models1 Jun
  17. 2022Pseudo Numerical Methods for Diffusion Models on Manifolds20 Feb
  18. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  19. 2021Tackling the Generative Learning Trilemma with Denoising Diffusion GANs15 Dec
  20. 2021Diffusion Models Beat GANs on Image Synthesis11 May
  21. 2021Knowledge Distillation in Iterative Generative Models for Improved Sampling Speed7 Jan
  22. 2020Score-Based Generative Modeling through Stochastic Differential Equations26 Nov
  23. 2020Denoising Diffusion Implicit Models6 Oct
  24. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 45

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  3. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  4. 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,585
  5. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,355
  6. 2023Adding Conditional Control to Text-to-Image Diffusion Modelscited by 5,503
  7. 2021Zero-Shot Text-to-Image Generationcited by 4,288
  8. 2022LAION-5B: An open large-scale dataset for training next generation image-text modelscited by 4,152
  9. 2021GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Modelscited by 3,614
  10. 2022DreamFusion: Text-to-3D using 2D Diffusioncited by 3,025
  11. 2018Large Scale GAN Training for High Fidelity Natural Image Synthesiscited by 2,626
  12. 2022InstructPix2Pix: Learning to Follow Image Editing Instructionscited by 2,611
33 more
  1. 2023Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasetscited by 2,581
  2. 2022Imagen Video: High Definition Video Generation with Diffusion Modelscited by 1,778
  3. 2022RePaint: Inpainting using Denoising Diffusion Probabilistic Modelscited by 1,430
  4. 2022Scaling Autoregressive Models for Content-Rich Text-to-Image Generationcited by 1,295
  5. 2023ProlificDreamer: High-Fidelity and Diverse Text-to-3D Generation with Variational Score Distillationcited by 1,219
  6. 2023ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generationcited by 1,045
  7. 2022eDiff-I: Text-to-Image Diffusion Models with an Ensemble of Expert Denoiserscited by 931
  8. 2019CNN-generated images are surprisingly easy to spot... for nowcited by 847
  9. 2018Multimodal Unsupervised Image-to-Image Translationcited by 795
  10. 2023Diffusion Model Alignment Using Direct Preference Optimizationcited by 790
  11. 2018Which Training Methods for GANs do actually Converge?cited by 679
  12. 2022Score Jacobian Chaining: Lifting Pretrained 2D Diffusion Models for 3D Generationcited by 597
  13. 2022StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasetscited by 492
  14. 2023Scaling up GANs for Text-to-Image Synthesiscited by 492
  15. 2022Make-A-Scene: Scene-Based Text-to-Image Generation with Human Priorscited by 487
  16. 2023InstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generationcited by 395
  17. 2023DIRE for Diffusion-Generated Image Detectioncited by 390
  18. 2023Emu Edit: Precise Image Editing via Recognition and Generation Taskscited by 324
  19. 2017Toward Multimodal Image-to-Image Translationcited by 296
  20. 2022SmartBrush: Text and Shape Guided Object Inpainting with Diffusion Modelcited by 278
  21. 2023StyleGAN-T: Unlocking the Power of GANs for Fast Large-Scale Text-to-Image Synthesiscited by 212
  22. 2021Projected GANs Converge Fastercited by 208
  23. 2022Diffusion-GAN: Training GANs with Diffusioncited by 205
  24. 2022Scalable Adaptive Computation for Iterative Generationcited by 195
  25. 2023UFOGen: You Forward Once Large Scale Text-to-Image Generation via Diffusion GANscited by 166
  26. 2021LAFITE: Towards Language-Free Training for Text-to-Image Generationcited by 128
  27. 2023Delta Denoising Scorecited by 121
  28. 2023SwiftBrush: One-Step Text-to-Image Diffusion Model with Variational Score Distillationcited by 110
  29. 2024PeRFlow: Piecewise Rectified Flow as Universal Plug-and-Play Acceleratorcited by 97
  30. 2023Finetuning Text-to-Image Diffusion Models for Fairnesscited by 92
  31. 2022Any-resolution Training for High-resolution Image Synthesiscited by 72
  32. 2023HiPA: Enabling One-Step Text-to-Image Diffusion Models via High-Frequency-Promoting Adaptationcited by 16
  33. 2023MonoFlow: Rethinking Divergence GANs via the Perspective of Wasserstein Gradient Flowscited by 13

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 52; 2025: 224; 2026: 408.

684 papers.

Sort:

By area:

  1. 2025Self Forcing: Bridging the Train-Test Gap in Autoregressive Video Diffusioncited by 625
  2. 2024From Slow Bidirectional to Fast Autoregressive Video Diffusion Modelscited by 433
  3. 2024One Step Diffusion via Shortcut Modelscited by 391
  4. 2024Simplifying, Stabilizing and Scaling Continuous-Time Consistency Modelscited by 283
  5. 2025LongLive: Real-time Interactive Long Video Generationcited by 265
  6. 2025Rolling Forcing: Autoregressive Long Video Diffusion in Real Timecited by 234
  7. 2025SkyReels-V2: Infinite-length Film Generative Modelcited by 229
  8. 2024Deep Compression Autoencoder for Efficient High-Resolution Diffusion Modelscited by 217
  9. 2025Sparse VideoGen: Accelerating Video Diffusion Transformers with Spatial-Temporal Sparsitycited by 217
  10. 2025Self-Forcing++: Towards Minute-Scale High-Quality Video Generationcited by 192
  11. 2024SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Modelscited by 170
  12. 2025WorldPlay: Towards Long-Term Geometric Consistency for Real-Time Interactive World Modelingcited by 170
38 more
  1. 2025Difix3D+: Improving 3D Reconstructions with Single-Step Diffusion Modelscited by 161
  2. 2026Advancing Open-source World Modelscited by 154
  3. 2025Diffusion Adversarial Post-Training for One-Step Video Generationcited by 151
  4. 2025Frame Context Packing and Drift Prevention in Next-Frame-Prediction Video Diffusion Modelscited by 143
  5. 2025HiDream-I1: A High-Efficient Image Generative Foundation Model with Sparse Diffusion Transformercited by 141
  6. 2026DreamDojo: A Generalist Robot World Model from Large-Scale Human Videoscited by 135
  7. 2025VSA: Faster Video Diffusion with Trainable Sparse Attentioncited by 104
  8. 2025RELIC: Interactive Video World Model with Long-Horizon Memorycited by 96
  9. 2025Autoregressive Adversarial Post-Training for Real-Time Interactive Video Generationcited by 90
  10. 2024TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolutioncited by 87
  11. 2025SANA-Sprint: One-Step Diffusion with Continuous-Time Consistency Distillationcited by 87
  12. 2025Align Your Flow: Scaling Continuous-Time Flow Map Distillationcited by 87
  13. 2024Arbitrary-steps Image Super-resolution via Diffusion Inversioncited by 86
  14. 2025Large Scale Diffusion Distillation via Score-Regularized Continuous-Time Consistencycited by 86
  15. 2025Infinity-RoPE: Action-Controllable Infinite Video Generation Emerges From Autoregressive Self-Rolloutcited by 86
  16. 2025Reward Forcing: Efficient Streaming Video Generation with Rewarded Distribution Matching Distillationcited by 86
  17. 2025Ovi: Twin Backbone Cross-Modal Fusion for Audio-Video Generationcited by 83
  18. 2025MotionStream: Real-Time Video Generation with Interactive Motion Controlscited by 76
  19. 2024Simpler Diffusion (SiD2): 1.5 FID on ImageNet512 with pixel-space diffusioncited by 75
  20. 2025Kling-Omni Technical Reportcited by 71
  21. 2024Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decodingcited by 66
  22. 2026Helios: Real Real-Time Long Video Generation Modelcited by 65
  23. 2025GigaWorld-0: World Models as Data Engine to Empower Embodied AIcited by 62
  24. 2024Rectified Diffusion: Straightness Is Not Your Need in Rectified Flowcited by 61
  25. 2025Hunyuan-GameCraft-2: Instruction-following Interactive Game World Modelcited by 61
  26. 2026Context Forcing: Consistent Autoregressive Video Generation with Long Contextcited by 59
  27. 2025Learning Few-Step Diffusion Models by Trajectory Distribution Matchingcited by 57
  28. 2024Flash Diffusion: Accelerating Any Conditional Diffusion Model for Few Steps Image Generationcited by 56
  29. 2024Meissonic: Revitalizing Masked Generative Transformers for Efficient High-Resolution Text-to-Image Synthesiscited by 56
  30. 2026Causal Forcing++: Scalable Few-Step Autoregressive Diffusion Distillation for Real-Time Interactive Video Generationcited by 56
  31. 2024Adversarial Diffusion Compression for Real-World Image Super-Resolutioncited by 52
  32. 2025End-to-End Training for Autoregressive Video Diffusion via Self-Resamplingcited by 52
  33. 2025SeedVR2: One-Step Video Restoration via Diffusion Adversarial Post-Trainingcited by 50
  34. 2025Distribution Matching Distillation Meets Reinforcement Learningcited by 50
  35. 2026WoVR: World Models as Reliable Simulators for Post-Training VLA Policies with RLcited by 50
  36. 2024Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Stepcited by 47
  37. 2026Solaris: Building a Multiplayer Video World Model in Minecraftcited by 47
  38. 2025Decoupled DMD: CFG Augmentation as the Spear, Distribution Matching as the Shieldcited by 45

50 most cited of 684 All 684 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 20261.x-Distill: Breaking the Diversity, Quality, and Efficiency Barrier in Distribution Matching Distillation
  2. 2024Regularized Distribution Matching Distillation for One-step Unpaired Image-to-Image Translation
  3. 2024Data-Efficient Generation for Dataset Distillation
  4. 2022Progressive Distillation for Fast Sampling of Diffusion Models
  5. 2024Unlocking Dataset Distillation with Diffusion Models
  6. 2026AdvDMD: Adversarial Reward Meets DMD For High-Quality Few-Step Generation
  7. 2024Diffusion Models Are Innate One-Step Generators
  8. 2025Taming Diffusion for Dataset Distillation with High Representativeness