mlsanityData to 7 Oct 2026

Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Steps

arXiv 2501.09732 ↗First posted 16 Jan 2025cs.CVcited by 225

On the map: LLM reasoning → Diffusion reasoning and test-time scaling

also in 1 more
Abstract (arXiv)

Generative models have made significant impacts across various domains, largely due to their ability to scale during training by increasing data, computational resources, and model size, a phenomenon characterized by the scaling laws. Recent research has begun to explore inference-time scaling behavior in Large Language Models (LLMs), revealing how performance can further improve with additional computation during inference. Unlike LLMs, diffusion models inherently possess the flexibility to adjust inference-time computation via the number of denoising steps, although the performance gains typically flatten after a few dozen. In this work, we explore the inference-time scaling behavior of diffusion models beyond increasing denoising steps and investigate how the generation performance can further improve with increased computation. Specifically, we consider a search problem aimed at identifying better noises for the diffusion sampling process. We structure the design space along two axes: the verifiers used to provide feedback, and the algorithms used to find better noise candidates. Through extensive experiments on class-conditioned and text-conditioned image generation benchmarks, our findings reveal that increasing inference-time compute leads to substantial improvements in the quality of samples generated by diffusion models, and with the complicated nature of images, combinations of the components in the framework can be specifically chosen to conform with different application scenario.

Key facts. Cited by 225 papers in the map (data through 7 Oct 2026). Placed in Diffusion reasoning and test-time scaling, one of 17 directions in LLM reasoning. Builds on 45 of its 99 references that are in the map. First posted on arXiv 16 Jan 2025.

Builds on

45 of 99 references.

References · 45

  1. 2023GPT-4 Technical Reportcited by 18,132
  2. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  3. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  4. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,764
  5. 2020Scaling Laws for Neural Language Modelscited by 6,411
  6. 2022Classifier-Free Diffusion Guidancecited by 5,717
  7. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
  8. 2013Auto-Encoding Variational Bayescited by 3,980
  9. 2023Let's Verify Step by Stepcited by 3,598
  10. 2022Training Compute-Optimal Large Language Modelscited by 2,927
  11. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,388
  12. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,928
33 more
  1. 2023Consistency Modelscited by 1,778
  2. 2016An overview of gradient descent optimization algorithmscited by 1,021
  3. 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 946
  4. 2023Training Diffusion Models with Reinforcement Learningcited by 939
  5. 2023Math-Shepherd: Verify and Reinforce LLMs Step-by-step without Human Annotationscited by 895
  6. 2016Training Deep Nets with Sublinear Memory Costcited by 825
  7. 2023Gemini: A Family of Highly Capable Multimodal Modelscited by 810
  8. 2024SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformerscited by 707
  9. 2024Movie Gen: A Cast of Media Foundation Modelscited by 614
  10. 2024PixArt-Sigma: Weak-to-Strong Training of Diffusion Transformer for 4K Text-to-Image Generationcited by 326
  11. 2022The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Modelscited by 312
  12. 2024Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Modelscited by 282
  13. 2024Monte Carlo Tree Search Boosts Reasoning via Iterative Preference Learningcited by 238
  14. 2024Adjoint Matching: Fine-tuning Flow and Diffusion Generative Models with Memoryless Stochastic Optimal Controlcited by 222
  15. 2024Stream of Search (SoS): Learning to Search in Languagecited by 141
  16. 2024D-Flow: Differentiating through Flows for Controlled Generationcited by 115
  17. 2024ReNO: Enhancing One-step Text-to-Image Models through Reward-based Noise Optimizationcited by 115
  18. 2024Golden Noise for Diffusion Models: A Learning Frameworkcited by 91
  19. 2024Process Reward Model with Q-Value Rankingscited by 78
  20. 2024DITTO: Diffusion Inference-Time T-Optimization for Music Generationcited by 73
  21. 2024Not All Noises Are Created Equally:Diffusion Noise Selection and Optimizationcited by 70
  22. 2024MJ-Bench: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation?cited by 64
  23. 2022Medical Diffusion: Denoising Diffusion Probabilistic Models for 3D Medical Image Generationcited by 54
  24. 2024A Noise is Worth Diffusion Guidancecited by 48
  25. 2024Revisiting Text-to-Image Evaluation with Gecko: On Metrics, Prompts, and Human Ratingscited by 44
  26. 2023If at First You Don't Succeed, Try, Try Again: Faithful Diffusion-based Text-to-Image Generation by Selectioncited by 40
  27. 2024Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Tracescited by 40
  28. 2023ArchiSound: Audio Generation with Diffusioncited by 20
  29. 2024EVALALIGN: Supervised Fine-Tuning Multimodal LLMs with Human-Aligned Data for Evaluating Text-to-Image Modelscited by 20
  30. 2024Multimodal Large Language Model is a Human-Aligned Annotator for Text-to-Image Generationcited by 15
  31. 2024Diffusion Rejection Samplingcited by 11
  32. 2024Confidence-aware Reward Optimization for Fine-tuning Text-to-Image Modelscited by 9
  33. 2024New Desiderata for Direct Preference Optimizationcited by 5

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 145; 2026: 80.

225 papers.

Sort:

By area:

  1. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Modelscited by 381
  2. 2025A Survey on Test-Time Scaling in Large Language Models: What, How, Where, and How Well?cited by 215
  3. 2025In-Context Edit: Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformercited by 199
  4. 2025Remasking Discrete Diffusion Models with Inference-Time Scalingcited by 181
  5. 2025A Survey of Reinforcement Learning for Large Reasoning Modelscited by 175
  6. 2025CoT-Valve: Length-Compressible Chain-of-Thought Tuningcited by 167
  7. 2025SANA 1.5: Efficient Scaling of Training-Time and Inference-Time Compute in Linear Diffusion Transformercited by 165
  8. 2025Open-Sora 2.0: Training a Commercial-Level Video Generation Model in 200kcited by 157
  9. 2025Lumina-Image 2.0: A Unified and Efficient Image Generative Frameworkcited by 93
  10. 2025Llasa: Scaling Train-Time and Inference-Time Compute for Llama-based Speech Synthesiscited by 69
  11. 2025RewardDance: Reward Scaling in Visual Generationcited by 63
  12. 2025Inner Thinking Transformer: Leveraging Dynamic Depth Scaling to Foster Adaptive Internal Thinkingcited by 56
38 more
  1. 2025From Reflection to Perfection: Scaling Inference-Time Optimization for Text-to-Image Diffusion Models via Reflection Tuningcited by 53
  2. 2026Scaling Text-to-Image Diffusion Transformers with Representation Autoencoderscited by 49
  3. 2025Inference-time Scaling of Diffusion Models through Classical Searchcited by 48
  4. 2025Reflect-DiT: Inference-Time Scaling for Text-to-Image Diffusion Transformers via In-Context Reflectioncited by 42
  5. 2025Inference-Time Text-to-Video Alignment with Diffusion Latent Beam Searchcited by 40
  6. 2025Video-T1: Test-Time Scaling for Video Generationcited by 39
  7. 2025RLVR-World: Training World Models with Reinforcement Learningcited by 38
  8. 2025Scaling Image and Video Generation via Test-Time Evolutionary Searchcited by 38
  9. 2025Inference-Time Scaling for Flow Models via Stochastic Generation and Rollover Budget Forcingcited by 37
  10. 2025Fine-Tuning Discrete Diffusion Models with Policy Gradient Methodscited by 36
  11. 2025Dynamic Search for Inference-Time Alignment in Diffusion Modelscited by 36
  12. 2025Lumina-mGPT 2.0: Stand-Alone AutoRegressive Image Modelingcited by 35
  13. 2025Noise Hypernetworks: Amortizing Test-Time Compute in Diffusion Modelscited by 35
  14. 2026Inference-time Physics Alignment of Video Generative Models with Latent World Modelscited by 32
  15. 2026Meta Flow Maps enable scalable reward alignmentcited by 32
  16. 2025Diffusion Tree Sampling: Scalable inference-time alignment of diffusion modelscited by 31
  17. 2025Energy-Based Transformers are Scalable Learners and Thinkerscited by 30
  18. 2025U-REPA: Aligning Diffusion U-Nets to ViTscited by 29
  19. 2025Learning Long-Context Diffusion Policies via Past-Token Predictioncited by 28
  20. 2025IMAGHarmony: Controllable Image Editing with Consistent Object Quantity and Layoutcited by 28
  21. 2025Test-Time Scaling of Diffusion Models via Noise Trajectory Searchcited by 28
  22. 2025Scaling Offline RL via Efficient and Expressive Shortcut Modelscited by 27
  23. 2026Diamond Maps: Efficient Reward Alignment via Stochastic Flow Mapscited by 26
  24. 2025Energy Matching: Unifying Flow Matching and Energy-Based Models for Generative Modelingcited by 25
  25. 2025textttComplex-Edit: CoT-Like Instruction Generation for Complexity-Controllable Image Editing Benchmarkcited by 25
  26. 2025VeriThinker: Learning to Verify Makes Reasoning Model Efficientcited by 24
  27. 2025Training-Free Guidance Beyond Differentiability: Scalable Path Steering with Tree Search in Diffusion and Flow Modelscited by 23
  28. 2025RePrompt: Reasoning-Augmented Reprompting for Text-to-Image Generation via Reinforcement Learningcited by 22
  29. 2025Cycle Consistency as Reward: Learning Image-Text Alignment without Human Preferencescited by 22
  30. 2025HiFlow: Training-free High-Resolution Image Generation with Flow-Aligned Guidancecited by 21
  31. 2025DriftLite: Lightweight Drift Control for Inference-Time Scaling of Diffusion Modelscited by 21
  32. 2025Decoupled MeanFlow: Turning Flow Models into Flow Maps for Accelerated Samplingcited by 19
  33. 2025Feedback Guidance of Diffusion Modelscited by 16
  34. 2025Test-time Prompt Refinement for Text-to-Image Modelscited by 16
  35. 2025RoVer: Robot Reward Model as Test-Time Verifier for Vision-Language-Action Modelcited by 16
  36. 2026GEMS: Agent-Native Multimodal Generation with Memory and Skillscited by 16
  37. 2025TTS-VAR: A Test-Time Scaling Framework for Visual Auto-Regressive Generationcited by 15
  38. 2025Test-time scaling of diffusions with flow mapscited by 15

50 most cited of 225 All 225 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Bigger is not Always Better: Scaling Properties of Latent Diffusion Models
  2. 2026S3: Stratified Scaling Search for Test-Time in Diffusion Language Models
  3. 2025Navigating the Exploration-Exploitation Tradeoff in Inference-Time Scaling of Diffusion Models
  4. 2026Flash-BoN: Instant Drafts for Inference-Time Scaling in Diffusion Models
  5. 2026Scaling Beyond Masked Diffusion Language Models
  6. 2025Fewer Denoising Steps or Cheaper Per-Step Inference: Towards Compute-Optimal Diffusion Model Deployment
  7. 2025Scaling Behavior of Discrete Diffusion Language Models
  8. 2026Not All Denoising Steps Are Equal: Model Scheduling for Faster Masked Diffusion Language Models