mlsanityData to 14 Oct 2026

Stochastic Self-Guidance for Training-Free Enhancement of Diffusion Models

arXiv 2508.12880 ↗First posted 18 Aug 2025cs.CVcited by 36

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

Classifier-free Guidance (CFG) is a widely used technique in modern diffusion models for enhancing sample quality and prompt adherence. However, through an empirical analysis on Gaussian mixture modeling with a closed-form solution, we observe a discrepancy between the suboptimal results produced by CFG and the ground truth. The model's excessive reliance on these suboptimal predictions often leads to semantic incoherence and low-quality outputs. To address this issue, we first empirically demonstrate that the model's suboptimal predictions can be effectively refined using sub-networks of the model itself. Building on this insight, we propose S2-Guidance, a novel method that leverages stochastic block-dropping during the forward process to construct stochastic sub-networks, effectively guiding the model away from potential low-quality predictions and toward high-quality outputs. Extensive qualitative and quantitative experiments on text-to-image and text-to-video generation tasks demonstrate that S2-Guidance delivers superior performance, consistently surpassing CFG and other advanced guidance strategies. Our code will be released.

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 69 of its 88 references that are in the map. First posted on arXiv 18 Aug 2025.

Builds on

69 of 88 references.

Closest ancestors · 28

Same direction, newest first

  1. 2025Accelerating Diffusion Model Training under Minimal Budgets: A Condensation-Based Perspective8 Jul
  2. 2025Angle Domain Guidance: Latent Diffusion Requires Rotation Rather Than Extrapolation21 May
  3. 2025CFG-Zero*: Improved Classifier-Free Guidance for Flow Matching Models24 Mar
  4. 2025USP: Unified Self-Supervised Pretraining for Image Generation and Understanding8 Mar
  5. 2024Spatiotemporal Skip Guidance for Enhanced Video Diffusion Sampling27 Nov
  6. 2024Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models3 Oct
  7. 2024Token Caching for Diffusion Transformer Acceleration27 Sep
  8. 2024Classifier-Free Guidance is a Predictor-Corrector16 Aug
  9. 2024Smoothed Energy Guidance: Guiding Diffusion Models with Reduced Energy Curvature of Attention1 Aug
  10. 2024Diffusion Models in Low-Level Vision: A Survey17 Jun
  11. 2024CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models12 Jun
  12. 2024Guiding a Diffusion Model with a Bad Version of Itself4 Jun
16 more
  1. 2024Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models11 Apr
  2. 2024Self-Rectifying Diffusion Sampling with Perturbed-Attention Guidance26 Mar
  3. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesis5 Mar
  4. 2024Convergence Analysis for General Probability Flow ODEs of Diffusion Models in Wasserstein Distances31 Jan
  5. 2024SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers16 Jan
  6. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis4 Jul
  7. 2023Masked Diffusion Models Are Fast Distribution Learners20 Jun
  8. 2022Scalable Diffusion Models with Transformers19 Dec
  9. 2022Improving Sample Quality of Diffusion Models Using Self-Attention Guidance3 Oct
  10. 2022Diffusion Models in Vision: A Survey10 Sep
  11. 2022Classifier-Free Diffusion Guidance26 Jul
  12. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  13. 2021Diffusion Models Beat GANs on Image Synthesis11 May
  14. 2020Score-Based Generative Modeling through Stochastic Differential Equations26 Nov
  15. 2020Denoising Diffusion Implicit Models6 Oct
  16. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 41

  1. 2015Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learningcited by 4,448
  2. 2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,926
  3. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,700
  4. 2025Wan: Open and Advanced Large-Scale Video Generative Modelscited by 2,708
  5. 2023VBench: Comprehensive Benchmark Suite for Video Generative Modelscited by 1,822
  6. 2024HunyuanVideo: A Systematic Framework For Large Video Generative Modelscited by 1,660
  7. 2023Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesiscited by 890
  8. 2021Twins: Revisiting the Design of Spatial Attention in Vision Transformerscited by 565
  9. 2023Q-Align: Teaching LMMs for Visual Scoring via Discrete Text-Defined Levelscited by 503
  10. 2023FateZero: Fusing Attentions for Zero-shot Text-based Video Editingcited by 466
  11. 2021Conditional Positional Encodings for Vision Transformerscited by 406
  12. 2020BatchEnsemble: An Alternative Approach to Efficient Ensemble and Lifelong Learningcited by 371
29 more
  1. 2017Snapshot Ensembles: Train 1, get M for freecited by 350
  2. 2021The Intrinsic Dimension of Images and Its Impact on Learningcited by 315
  3. 2023Follow Your Pose: Pose-Guided Text-to-Video Generation using Pose-Free Videoscited by 314
  4. 2024Taming Rectified Flow for Inversion and Editingcited by 204
  5. 2024Follow-Your-Emoji: Fine-Controllable and Expressive Freestyle Portrait Animationcited by 174
  6. 2025GPG: A Simple and Strong Reinforcement Learning Baseline for Model Reasoningcited by 150
  7. 2024DiTFastAttn: Attention Compression for Diffusion Transformer Modelscited by 115
  8. 2024Stable Flow: Vital Layers for Training-Free Image Editingcited by 113
  9. 2025Follow-Your-Motion: Video Motion Transfer via Efficient Spatial-Temporal Decoupled Finetuningcited by 81
  10. 2022Verifying the Union of Manifolds Hypothesis for Image Datacited by 64
  11. 2024COVE: Unleashing the Diffusion Feature Correspondence for Consistent Video Editingcited by 59
  12. 2025VMBench: A Benchmark for Perception-Aligned Video Motion Generationcited by 49
  13. 2022DiffStyler: Controllable Dual Diffusion for Text-Driven Image Stylizationcited by 45
  14. 2025Understanding the Repeat Curse in Large Language Models from a Feature Perspectivecited by 37
  15. 2024InstantSwap: Fast Customized Concept Swapping across Sharp Shape Differencescited by 34
  16. 2024VisionLLaMA: A Unified LLaMA Backbone for Vision Taskscited by 28
  17. 2023LIME: Localized Image Editing via Attention Regularization in Diffusion Modelscited by 27
  18. 2025Omni-Effects: Unified and Spatially-Controllable Visual Effects Generationcited by 27
  19. 2024Real-world Image Dehazing with Coherence-based Pseudo Labeling and Cooperative Unfolding Networkcited by 26
  20. 2023Region-Aware Diffusion for Zero-shot Text-driven Image Editingcited by 25
  21. 2025FingER: Content Aware Fine-grained Evaluation with Reasoning for AI-Generated Videoscited by 22
  22. 2025NarrLV: Towards a Comprehensive Narrative-Centric Evaluation for Long Video Generationcited by 22
  23. 2023Efficient Robustness Assessment via Adversarial Spatial-Temporal Focus on Videoscited by 20
  24. 2024A Survey of AI-Generated Video Evaluationcited by 17
  25. 2025Integrating Extra Modality Helps Segmentor Find Camouflaged Objects Wellcited by 14
  26. 2025Safe-Sora: Safe Text-to-Video Generation via Graphical Watermarkingcited by 11
  27. 2025Fast Adversarial Training with Weak-to-Strong Spatial-Temporal Consistency in the Frequency Domain on Videoscited by 9
  28. 2024Revealing the Dark Secrets of Extremely Large Kernel ConvNets on Robustnesscited by 8
  29. 2025SPG: Improving Motion Diffusion by Smooth Perturbation Guidancecited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 10; 2026: 26.

36 papers.

Sort:

By area:

  1. 2025World4RL: Diffusion World Models for Policy Refinement with Reinforcement Learning for Robotic Manipulationcited by 35
  2. 2025There is No VAE: End-to-End Pixel-Space Generative Modeling via Self-Supervised Pre-trainingcited by 24
  3. 2026FastVMT: Eliminating Redundancy in Video Motion Transfercited by 23
  4. 2025Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenescited by 20
  5. 2026Generation Enhances Understanding in Unified Multimodal Models via Multi-Representation Generationcited by 20
  6. 2025Taming Preference Mode Collapse via Directional Decoupling Alignment in Diffusion Reinforcement Learningcited by 18
  7. 2025Guiding a Diffusion Transformer with the Internal Dynamics of Itselfcited by 16
  8. 2026Group Editing: Edit Multiple Images in One Gocited by 11
  9. 2025ImagerySearch: Adaptive Test-Time Search for Video Generation Beyond Semantic Dependency Constraintscited by 9
  10. 2025MACE-Dance: Motion-Appearance Cascaded Experts for Music-Driven Dance Video Generationcited by 9
  11. 2026From Scale to Speed: Adaptive Test-Time Scaling for Image Editingcited by 8
  12. 2026PreciseCache: Precise Feature Caching for Efficient and High-fidelity Video Generationcited by 8
24 more
  1. 2026C2FG: Control Classifier-Free Guidance via Score Discrepancy Analysiscited by 7
  2. 2026Photon: Speedup Volume Understanding with Efficient Multimodal Large Language Modelscited by 6
  3. 2026Embedding-perturbed Exploration Preference Optimization for Flow Modelscited by 5
  4. 2025Taming Hallucinations: Boosting MLLMs' Video Understanding via Counterfactual Video Generationcited by 4
  5. 2026Guidance Matters: Rethinking the Evaluation Pitfall for Text-to-Image Generationcited by 4
  6. 2026One Model, Many Budgets: Elastic Latent Interfaces for Diffusion Transformerscited by 4
  7. 2026Artifact-Aware Evaluation for High-Quality Video Generationcited by 3
  8. 2026Exposing and Defending the Achilles' Heel of Video Mixture-of-Expertscited by 3
  9. 2026Elucidating the SNR-t Bias of Diffusion Probabilistic Modelscited by 3
  10. 2026OmniDance: Multimodal Driven Dance Video Generation with Large-scale Internet Datacited by 3
  11. 2026Steering Video Diffusion Transformers with Massive Activationscited by 2
  12. 2026ConceptWeaver: Weaving Disentangled Concepts with Flowcited by 2
  13. 2026Extending One-Step Image Generation from Class Labels to Text via Discriminative Text Representationcited by 2
  14. 2026GuidedBridge: Training-freely Improving Bridge Models with Prior Guidancecited by 2
  15. 2026Persistence Forcing: Exploiting Feature Specialization in Pixel-Space Diffusioncited by 2
  16. 2026MoKus: Leveraging Cross-Modal Knowledge Transfer for Knowledge-Aware Concept Customizationcited by 1
  17. 2026Improving Diffusion Generalization with Weak-to-Strong Segmented Guidancecited by 1
  18. 2026InstanceAnimator: Multi-Instance Sketch Video Colorizationcited by 1
  19. 2026Enhancing Train-Free Infinite-Frame Generation for Consistent Long Videoscited by 1
  20. 2025Home-made Diffusion Model from Scratch to Hatchcited by 0
  21. 2025EMAG: Self-Rectifying Diffusion Sampling with Exponential Moving Average Guidancecited by 0
  22. 2026Awakening Diffusion Transformers: Eliciting Stronger Generation and Understanding via Massive Activation Modulationcited by 0
  23. 2026Accelerating Diffusion Transformers with Gaussian Process Rectified Feature Cachecited by 0
  24. 2026Images Talk Back: Guiding MMDiTs by Suppressing Image-to-Text Feedbackcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Revisiting Classifier-Free Guidance Methods in Latent Diffusion Models
  2. 2023Bridging the Gap: Addressing Discrepancies in Diffusion Model Training for Classifier-Free Guidance
  3. 2025Improving Classifier-Free Guidance in Masked Diffusion: Low-Dim Theoretical Insights with High-Dim Impact
  4. 2024No Training, No Problem: Rethinking Classifier-Free Guidance for Diffusion Models
  5. 2025Feedback Guidance of Diffusion Models
  6. 2025Learn to Guide Your Diffusion Model
  7. 2023Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models
  8. 2025Diffusion Models without Classifier-free Guidance