mlsanityData to 7 Oct 2026

Classifier-Free Guidance is a Predictor-Corrector

arXiv 2408.09000 ↗First posted 16 Aug 2024cs.LG cs.AI cs.CVcited by 79

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

We investigate the theoretical foundations of classifier-free guidance (CFG). CFG is the dominant method of conditional sampling for text-to-image diffusion models, yet unlike other aspects of diffusion, it remains on shaky theoretical footing. In this paper, we disprove common misconceptions, by showing that CFG interacts differently with DDPM (Ho et al., 2020) and DDIM (Song et al., 2021), and neither sampler with CFG generates the gamma-powered distribution p(x|c)γ p(x)1-γ. Then, we clarify the behavior of CFG by showing that it is a kind of predictor-corrector method (Song et al., 2020) that alternates between denoising and sharpening, which we call predictor-corrector guidance (PCG). We prove that in the SDE limit, CFG is actually equivalent to combining a DDIM predictor for the conditional distribution together with a Langevin dynamics corrector for a gamma-powered distribution (with a carefully chosen gamma). Our work thus provides a lens to theoretically understand CFG by embedding it in a broader design space of principled sampling methods.

Key facts. Cited by 79 papers in the map (data through 7 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 25 of its 47 references that are in the map. First posted on arXiv 16 Aug 2024.

Builds on

25 of 47 references.

Closest ancestors · 18

Same direction, newest first

  1. 2024Guiding a Diffusion Model with a Bad Version of Itself4 Jun
  2. 2024Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models11 Apr
  3. 2024Theoretical Insights for Diffusion Guidance: A Case Study for Gaussian Mixture Models3 Mar
  4. 2023Analyzing and Improving the Training Dynamics of Diffusion Models5 Dec
  5. 2023SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis4 Jul
  6. 2023Reduce, Reuse, Recycle: Compositional Generation with Energy-Based Diffusion Models and MCMC22 Feb
  7. 2023Universal Guidance for Diffusion Models14 Feb
  8. 2022DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models2 Nov
  9. 2022Sampling is as easy as learning the score: theory for diffusion models with minimal data assumptions22 Sep
  10. 2022Classifier-Free Diffusion Guidance26 Jul
  11. 2022DPM-Solver: A Fast ODE Solver for Diffusion Probabilistic Model Sampling in Around 10 Steps2 Jun
  12. 2022Elucidating the Design Space of Diffusion-Based Generative Models1 Jun
6 more
  1. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  2. 2021Gotta Go Fast When Generating Data with Score-Based Models28 May
  3. 2021Diffusion Models Beat GANs on Image Synthesis11 May
  4. 2020Score-Based Generative Modeling through Stochastic Differential Equations26 Nov
  5. 2020Denoising Diffusion Implicit Models6 Oct
  6. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 7

  1. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  2. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  3. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,067
  4. 2021GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Modelscited by 3,595
  5. 2022RePaint: Inpainting using Denoising Diffusion Probabilistic Modelscited by 1,419
  6. 2023Single-Stage Diffusion NeRF: A Unified Approach to 3D Generation and Reconstructioncited by 180
  7. 2015A Complete Recipe for Stochastic Gradient MCMCcited by 146

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 11; 2025: 40; 2026: 28.

79 papers.

Sort:

By area:

  1. 2025Feynman-Kac Correctors in Diffusion: Annealing, Guidance, and Product of Expertscited by 85
  2. 2025CFG-Zero*: Improved Classifier-Free Guidance for Flow Matching Modelscited by 47
  3. 2024What does guidance do? A fine-grained analysis in a simple settingcited by 45
  4. 2025UniEdit-Flow: Unleashing Inversion and Editing in the Era of Flow Modelscited by 38
  5. 2024The Superposition of Diffusion Models Using the Ito Density Estimatorcited by 36
  6. 2025Stochastic Self-Guidance for Training-Free Enhancement of Diffusion Modelscited by 35
  7. 2026Q-learning with Adjoint Matchingcited by 33
  8. 2025Debiasing Guidance for Discrete Diffusion with Sequential Monte Carlocited by 29
  9. 2024Shielded Diffusion: Generating Novel and Diverse Images using Sparse Repellencycited by 21
  10. 20243D-Adapter: Geometry-Consistent Multi-View Diffusion for High-Quality 3D Generationcited by 19
  11. 2024Ensemble Kalman Diffusion Guidance: A Derivative-free Method for Inverse Problemscited by 18
  12. 2024Rectified Diffusion Guidance for Conditional Generationcited by 18
38 more
  1. 2025UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Controlcited by 18
  2. 2025Composition and Control with Distilled Energy Diffusion Models and Sequential Monte Carlocited by 18
  3. 2025Solving Inverse Problems via Diffusion-Based Priors: An Approximation-Free Ensemble Sampling Approachcited by 18
  4. 2025Gaussian Mixture Flow Matching Modelscited by 16
  5. 2025Feedback Guidance of Diffusion Modelscited by 16
  6. 2026Qwen-Image-Bench: From Generation to Creation in Text-to-Image Evaluationcited by 16
  7. 2025Stage-wise Dynamics of Classifier-Free Guidance in Diffusion Modelscited by 13
  8. 2025Flow Matching with Semidiscrete Couplingscited by 12
  9. 2025Learn to Guide Your Diffusion Modelcited by 12
  10. 2025Improving Classifier-Free Guidance in Masked Diffusion: Low-Dim Theoretical Insights with High-Dim Impactcited by 11
  11. 2024Provable Convergence and Limitations of Geometric Tempering for Langevin Dynamicscited by 10
  12. 2025Adaptive Diffusion Guidance via Stochastic Optimal Controlcited by 9
  13. 2025What Exactly Does Guidance Do in Masked Discrete Diffusion Modelscited by 9
  14. 2025Provable Efficiency of Guidance in Diffusion Models for General Data Distributioncited by 8
  15. 2025Angle Domain Guidance: Latent Diffusion Requires Rotation Rather Than Extrapolationcited by 8
  16. 2025Composition and Alignment of Diffusion Models using Constrained Learningcited by 8
  17. 2025REG: Rectified Gradient Guidance for Conditional Diffusion Modelscited by 7
  18. 2025Inference-Time Alignment Control for Diffusion Models with Reinforcement Learning Guidancecited by 7
  19. 2025Towards a unified framework for guided diffusion modelscited by 7
  20. 2024Theory on Score-Mismatched Diffusion Models and Zero-Shot Conditional Samplerscited by 6
  21. 2025Self-Reflective Reinforcement Learning for Diffusion-based Image Reasoning Generationcited by 6
  22. 2025Towards a Golden Classifier-Free Guidance Path via Foresight Fixed Point Iterationscited by 6
  23. 2025AudioMoG: Guiding Audio Generation with Mixture-of-Guidancecited by 5
  24. 2026Continuous Control of Editing Models via Adaptive-Origin Guidancecited by 5
  25. 2025Conditional Diffusion Models with Classifier-Free Gibbs-like Guidancecited by 4
  26. 2026C2FG: Control Classifier-Free Guidance via Score Discrepancy Analysiscited by 4
  27. 2025Guidance Free Image Editing via Explicit Conditioningcited by 3
  28. 2026Emergence of Distortions in High-Dimensional Guided Diffusion Modelscited by 3
  29. 2026Composition of Memory Experts for Diffusion World Modelscited by 3
  30. 2024Feature-guided score diffusion for sampling conditional densitiescited by 2
  31. 2024Guiding a diffusion model using sliding windowscited by 2
  32. 2024A solvable generative model with a linear, one-step denoisercited by 2
  33. 2025DICE: Distilling Classifier-Free Guidance into Text Embeddingscited by 2
  34. 2025Towards Flow-Matching-based TTS without Classifier-Free Guidancecited by 2
  35. 2025On the Separability of Information in Diffusion Modelscited by 2
  36. 2025Tortoise and Hare Guidance: Accelerating Diffusion Model Inference with Multirate Integrationcited by 2
  37. 2025Progress by Pieces: Test-Time Scaling for Autoregressive Image Generationcited by 2
  38. 2026SceneNAT: Masked Generative Modeling for Language-Guided Indoor Scene Synthesiscited by 2

50 most cited of 79 All 79 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Classifier-free Guidance with Adaptive Scaling
  2. 2024EP-CFG: Energy-Preserving Classifier-Free Guidance
  3. 2025CFG-OEC: Classifier Free Guidance with Orthogonal Error Correction
  4. 2023Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models
  5. 2025Classifier-Free Guidance: From High-Dimensional Analysis to Generalized Guidance Forms
  6. 2026Guiding a Diffusion Model by Swapping Its Tokens
  7. 2026Revisiting Classifier-Free Guidance Methods in Latent Diffusion Models
  8. 2025Towards Understanding the Mechanisms of Classifier-Free Guidance