mlsanityData to 14 Oct 2026

DiFA: Inference-Time Forward-Process Alignment for Diffusion Models

arXiv 2607.17972 ↗First posted 20 Jul 2026cs.LGcited by 1

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

The prevailing inference framework for diffusion models formulates generation fundamentally as a problem of numerical integration. This perspective casts the model as an exact estimator, neglecting the inherent statistical uncertainty of the denoising process. In this work, we propose Forward-Process Aligned Diffusion prediction (DiFA), a training-free framework that reframes inference-time data prediction refinement as a sequential state estimation problem. Rather than reusing past outputs solely for numerical integration, DiFA treats iterative data predictions along the reverse trajectory as correlated observations to build a forward-aligned temporal consensus. Inspired by Kalman filtering, this consensus aggregates historical predictions according to structural consistency and noise-level compatibility. To counteract the over-smoothing tendency of temporal consensus, we introduce a deviation guidance mechanism to adaptively preserve residual details. Empirically, DiFA yields significant improvements on CIFAR-10 and ImageNet across the evaluated metrics, including FID, IS, and FD-DINOv2, demonstrating that aligning inference with the forward statistical structure substantially improves generative fidelity.

Key facts. Cited by 1 paper 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 20 Jul 2026.

Builds on

81 of 91 references.

Closest ancestors · 56

Same direction, newest first

  1. 2026Mitigating the Contractivity Trap in Diffusion ODEs via Stein Stabilization5 Jun
  2. 2025Information Theoretic Learning for Diffusion Models with Warm Start23 Oct
  3. 2025Temporal Alignment Guidance: On-Manifold Sampling in Diffusion Models13 Oct
  4. 2025Diffusion Models and the Manifold Hypothesis: Log-Domain Smoothing is Geometry Adaptive2 Oct
  5. 2025EVODiff: Entropy-aware Variance Optimized Diffusion Inference30 Sep
  6. 2025Antithetic Noise in Diffusion Models6 Jun
  7. 2025Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Steps16 Jan
  8. 2024Enhancing Sample Generation of Diffusion Models using Noise Level Correction7 Dec
  9. 2024One-Step Diffusion Distillation through Score Implicit Matching22 Oct
  10. 2024Adversarial Score identity Distillation: Rapidly Surpassing the Teacher in One Step19 Oct
  11. 2024One Step Diffusion via Shortcut Models16 Oct
  12. 2024TFG: Unified Training-Free Guidance for Diffusion Models24 Sep
38 more
  1. 2024Sparse Inducing Points in Deep Gaussian Processes: Enhancing Modeling with Denoising Diffusion Variational Inference24 Jul
  2. 2024DisCo-Diff: Enhancing Continuous Diffusion Models with Discrete Latents3 Jul
  3. 2024Guiding a Diffusion Model with a Bad Version of Itself4 Jun
  4. 2024Learning to Discretize Denoising Diffusion ODEs24 May
  5. 2024Align Your Steps: Optimizing Sampling Schedules in Diffusion Models22 Apr
  6. 2024Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance8 Apr
  7. 2024Self-Rectifying Diffusion Sampling with Perturbed-Attention Guidance26 Mar
  8. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesis5 Mar
  9. 2024Guidance with Spherical Gaussian Constraint for Conditional Diffusion5 Feb
  10. 2023Cache Me if You Can: Accelerating Diffusion Models through Block Caching6 Dec
  11. 2023Analyzing and Improving the Training Dynamics of Diffusion Models5 Dec
  12. 2023DeepCache: Accelerating Diffusion Models for Free1 Dec
  13. 2023One-step Diffusion with Distribution Matching Distillation30 Nov
  14. 2023DPM-Solver-v3: Improved Diffusion ODE Solver with Empirical Model Statistics20 Oct
  15. 2023Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inference6 Oct
  16. 2023Consistency Trajectory Models: Learning Probability Flow ODE Trajectory of Diffusion1 Oct
  17. 2023Elucidating the Exposure Bias in Diffusion Models29 Aug
  18. 2023SciRE-Solver: Accelerating Diffusion Models Sampling by Score-integrand Solver with Recursive Difference15 Aug
  19. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis4 Jul
  20. 2023Fast Diffusion Model12 Jun
  21. 2023On the Design Fundamentals of Diffusion Models: A Survey7 Jun
  22. 2023Parallel Sampling of Diffusion Models25 May
  23. 2023SEEDS: Exponential SDE Solvers for Fast High-Quality Sampling from Diffusion Models23 May
  24. 2023Common Diffusion Noise Schedules and Sample Steps are Flawed15 May
  25. 2023Consistency Models2 Mar
  26. 2023Understanding Diffusion Objectives as the ELBO with Simple Data Augmentation1 Mar
  27. 2023Universal Guidance for Diffusion Models14 Feb
  28. 2023UniPC: A Unified Predictor-Corrector Framework for Fast Sampling of Diffusion Models9 Feb
  29. 2022DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models2 Nov
  30. 2022On Distillation of Guided Diffusion Models6 Oct
  31. 2022Improving Sample Quality of Diffusion Models Using Self-Attention Guidance3 Oct
  32. 2022DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps2 Jun
  33. 2022Elucidating the Design Space of Diffusion-Based Generative Models1 Jun
  34. 2022Fast Sampling of Diffusion Models with Exponential Integrator29 Apr
  35. 2022Pseudo Numerical Methods for Diffusion Models on Manifolds20 Feb
  36. 2022Analytic-DPM: an Analytic Estimate of the Optimal Reverse Variance in Diffusion Probabilistic Models17 Jan
  37. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  38. 2021Variational Diffusion Models1 Jul

50 newest of 56 All 56 (JSON)

Other references · 25

  1. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,058
  2. 2022Flow Matching for Generative Modelingcited by 5,702
  3. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,700
  4. 2016Improved Techniques for Training GANscited by 2,970
  5. 2023Align your Latents: High-Resolution Video Synthesis with Latent Diffusion Modelscited by 1,476
  6. 2020Training Generative Adversarial Networks with Limited Datacited by 1,186
  7. 2025FLUX.1 Kontext: Flow Matching for In-Context Image Generation and Editing in Latent Spacecited by 1,006
  8. 2025Mean Flows for One-step Generative Modelingcited by 561
  9. 2022Improving Diffusion Models for Inverse Problems using Manifold Constraintscited by 508
  10. 2022StyleGAN-XL: Scaling StyleGAN to Large Diverse Datasetscited by 492
  11. 2023InstaFlow: One Step is Enough for High-Quality Diffusion-Based Text-to-Image Generationcited by 395
  12. 2023Diffusion Self-Guidance for Controllable Image Generationcited by 305
13 more
  1. 2021More Control for Free! Image Synthesis with Semantic Diffusion Guidancecited by 212
  2. 2023Exposing flaws of generative model evaluation metrics and their unfair treatment of diffusion modelscited by 205
  3. 2022Scalable Adaptive Computation for Iterative Generationcited by 195
  4. 2025Align Your Flow: Scaling Continuous-Time Flow Map Distillationcited by 87
  5. 2024Rectified Diffusion: Straightness Is Not Your Need in Rectified Flowcited by 61
  6. 2025The GAN is dead; long live the GAN! A Modern GAN Baselinecited by 59
  7. 2024Inference-Time Alignment of Diffusion Models with Direct Noise Optimizationcited by 55
  8. 2024Zigzag Diffusion Sampling: Diffusion Models Can Self-Improve via Self-Reflectioncited by 41
  9. 2024How to Continually Adapt Text-to-Image Diffusion Models for Flexible Customization?cited by 33
  10. 2024Coupled Mamba: Enhanced Multi-modal Fusion with Coupled State Space Modelcited by 31
  11. 2025LoRA-Mixer: Coordinate Modular LoRA Experts Through Serial Attention Routingcited by 18
  12. 2025Dequantified Diffusion-Schr"odinger Bridge for Density Ratio Estimationcited by 11
  13. 2024Entropy-Informed Weighting Channel Normalizing Flow for Deep Generative Modelscited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Improving Few-Step Language Flows with Untied Self-Conditioningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025ETC: training-free diffusion models acceleration with Error-aware Trend Consistency
  2. 2026Inference-Time Attribute Distribution Alignment for Unconditional Diffusion
  3. 2026Inference-Time Alignment for Diffusion Models via Variationally Stable Doob's Matching
  4. 2025RNE: plug-and-play diffusion inference-time control and energy-based training
  5. 2025DiffPro: Joint Timestep and Layer-Wise Precision Optimization for Efficient Diffusion Inference
  6. 2025Inference-Time Alignment of Diffusion Models via Evolutionary Algorithms
  7. 2026Diversify Diffusion with Temperature Sampling and Variance-Corrective Time Shifting
  8. 2025Dynamic Search for Inference-Time Alignment in Diffusion Models