mlsanityData to 14 Oct 2026

CFG-Zero*: Improved Classifier-Free Guidance for Flow Matching Models

arXiv 2503.18886 ↗First posted 24 Mar 2025cs.CVcited by 52

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

Classifier-Free Guidance (CFG) is a widely adopted technique in diffusion/flow models to improve image fidelity and controllability. In this work, we first analytically study the effect of CFG on flow matching models trained on Gaussian mixtures where the ground-truth flow can be derived. We observe that in the early stages of training, when the flow estimation is inaccurate, CFG directs samples toward incorrect trajectories. Building on this observation, we propose CFG-Zero*, an improved CFG with two contributions: (a) optimized scale, where a scalar is optimized to correct for the inaccuracies in the estimated velocity, hence the * in the name; and (b) zero-init, which involves zeroing out the first few steps of the ODE solver. Experiments on both text-to-image (Lumina-Next, Stable Diffusion 3, and Flux) and text-to-video (Wan-2.1) generation demonstrate that CFG-Zero* consistently outperforms CFG, highlighting its effectiveness in guiding Flow Matching models. (Code is available at github.com/WeichenFan/CFG-Zero-star)

Key facts. Cited by 52 papers in the map (data through 14 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 34 of its 43 references that are in the map. First posted on arXiv 24 Mar 2025.

Builds on

34 of 43 references.

Closest ancestors · 19

Same direction, newest first

  1. 2024Rectified Diffusion Guidance for Conditional Generation24 Oct
  2. 2024Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models3 Oct
  3. 2024Classifier-Free Guidance is a Predictor-Corrector16 Aug
  4. 2024CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models12 Jun
  5. 2024Lumina-Next: Making Lumina-T2X Stronger and Faster with Next-DiT5 Jun
  6. 2024Guiding a Diffusion Model with a Bad Version of Itself4 Jun
  7. 2024Analysis of Classifier-Free Guidance Weight Schedulers19 Apr
  8. 2024Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models11 Apr
  9. 2024Scaling Rectified Flow Transformers for High-Resolution Image Synthesis5 Mar
  10. 2024SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers16 Jan
  11. 2023Characteristic Guidance: Non-linear Correction for Diffusion Model at Large Guidance Scale11 Dec
  12. 2022Scalable Diffusion Models with Transformers19 Dec
7 more
  1. 2022Classifier-Free Diffusion Guidance26 Jul
  2. 2022Elucidating the Design Space of Diffusion-Based Generative Models1 Jun
  3. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  4. 2021Diffusion Models Beat GANs on Image Synthesis11 May
  5. 2020Score-Based Generative Modeling through Stochastic Differential Equations26 Nov
  6. 2020Denoising Diffusion Implicit Models6 Oct
  7. 2020Denoising Diffusion Probabilistic Models19 Jun

Other references · 15

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2022Flow Matching for Generative Modelingcited by 5,702
  3. 2022Generative Adversarial Networkscited by 5,627
  4. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,101
  5. 2022Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flowcited by 3,700
  6. 2016Improved Techniques for Training GANscited by 2,970
  7. 2024CogVideoX: Text-to-Video Diffusion Models with An Expert Transformercited by 2,261
  8. 2023VBench: Comprehensive Benchmark Suite for Video Generative Modelscited by 1,822
  9. 2024HunyuanVideo: A Systematic Framework For Large Video Generative Modelscited by 1,660
  10. 2017NIMA: Neural Image Assessmentcited by 389
  11. 2024Flow Matching Guide and Codecited by 301
  12. 2023T2I-CompBench++: An Enhanced and Comprehensive Benchmark for Compositional Text-to-image Generationcited by 260
3 more
  1. 2023Guided Flows for Generative Modeling and Decision Makingcited by 101
  2. 2019Sliced Wasserstein Generative Modelscited by 61
  3. 2025Vchitect-2.0: Parallel Transformer for Scaling Up Video Diffusion Modelscited by 60

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 17; 2026: 35.

52 papers.

Sort:

By area:

  1. 2025Latent Diffusion Model without Variational Autoencodercited by 88
  2. 2025LongCat-Video Technical Reportcited by 85
  3. 2026Helios: Real Real-Time Long Video Generation Modelcited by 65
  4. 2025Stochastic Self-Guidance for Training-Free Enhancement of Diffusion Modelscited by 36
  5. 2025Rectified-CFG++ for Flow Based Modelscited by 17
  6. 2025Guiding a Diffusion Transformer with the Internal Dynamics of Itselfcited by 16
  7. 2026Self-Refining Video Samplingcited by 15
  8. 2025Learn to Guide Your Diffusion Modelcited by 14
  9. 2025Normalized Attention Guidance: Universal Negative Guidance for Diffusion Modelscited by 13
  10. 2025FlowerDance: MeanFlow for Efficient and Refined 3D Dance Generationcited by 11
  11. 2025Inference-Time Alignment Control for Diffusion Models with Reinforcement Learning Guidancecited by 10
  12. 2026CFG-Ctrl: Control-Based Classifier-Free Diffusion Guidancecited by 8
38 more
  1. 2025Pyramidal Patchification Flow for Visual Generationcited by 4
  2. 2025Selective Classifier-free Guidance for Zero-shot Text-to-speechcited by 4
  3. 2026Training-Free Refinement of Flow Matching with Divergence-based Samplingcited by 4
  4. 2026Delta Score Matters! Spatial Adaptive Multi Guidance in Diffusion Modelscited by 4
  5. 2025SAD-Flower: Flow Matching for Safe, Admissible, and Dynamically Consistent Planningcited by 3
  6. 2026Improving Classifier-Free Guidance of Flow Matching via Manifold Projectioncited by 3
  7. 2025Towards Flow-Matching-based TTS without Classifier-Free Guidancecited by 2
  8. 2026Free Lunch for Stabilizing Rectified Flow Inversioncited by 2
  9. 2026Steering Video Diffusion Transformers with Massive Activationscited by 2
  10. 2026ConceptWeaver: Weaving Disentangled Concepts with Flowcited by 2
  11. 2026GenMed: A Pairwise Generative Reformulation of Medical Diagnostic Taskscited by 2
  12. 2026AssetGen: Deployable 3D Asset Generation at Interactive Speedcited by 2
  13. 2026Off-Manifold Refinement: Guiding Video Generators with a Frozen World Modelcited by 2
  14. 2025MAMBO-G: Magnitude-Aware Mitigation for Boosted Guidancecited by 1
  15. 2025SAIP: A Plug-and-Play Scale-adaptive Module in Diffusion-based Inverse Problemscited by 1
  16. 2026Improving Diffusion Generalization with Weak-to-Strong Segmented Guidancecited by 1
  17. 2026VAGS: Velocity Adaptive Guidance Scale for Image Editing and Generationcited by 1
  18. 2026SpecLoR: Spectral Lookahead Rectification for Motion-Coherent Text-to-Video Generationcited by 1
  19. 2026ModaFlow: Modality-Aware Flow Matching for High-Fidelity Virtual Try-Oncited by 1
  20. 2026PathGuide: Dynamic Classifier-Free Guidance via On-Policy Transport Alignmentcited by 1
  21. 2025MeanFlow-Accelerated Multimodal Video-to-Audio Synthesis via One-Step Generationcited by 0
  22. 2025CFG-OEC: Classifier Free Guidance with Orthogonal Error Correctioncited by 0
  23. 2026Single-step Controllable Music Bandwidth Extension With Flow Matchingcited by 0
  24. 2026Rethinking Global Text Conditioning in Diffusion Transformerscited by 0
  25. 2026Reflective Flow Sampling Enhancementcited by 0
  26. 2026When to Lock Attention: Training-Free KV Control in Video Diffusioncited by 0
  27. 2026Exploring Motion-Language Alignment for Text-driven Motion Generationcited by 0
  28. 2026Awakening Diffusion Transformers: Eliciting Stronger Generation and Understanding via Massive Activation Modulationcited by 0
  29. 2026Guidance Breaks the Fitted Operator: A Terminal-Fitted Repair for Classifier-Free Guidancecited by 0
  30. 2026FreyaTTS: A Compact Tokenizer-Free Flow-Matching Transformer for Turkish-First Speech Synthesiscited by 0
  31. 2026Lyapunov Guidance: A Unified Framework for Stabilizing Generative Flowscited by 0
  32. 2026BridgeFlow: Fast and Robust SE(2)-Equivariant Motion Planning with Flow Matchingcited by 0
  33. 2026Signed Rectified Flow: Negativity-Controlled Generationcited by 0
  34. 2026Revisiting Classifier-Free Guidance Methods in Latent Diffusion Modelscited by 0
  35. 2026SAGE: Subspace Alignment for Classifier-Free Guidance in Mixture-of-Experts Diffusion Modelscited by 0
  36. 2026UltraDub: Towards Authentic Dubbing by Unifying Visually-Steered Flow Learning and Trajectory Guidancecited by 0
  37. 2026AnchorGen: Anchored Optimization for Customizable Generative 3D Designcited by 0
  38. 2026Global Transport Couplings for Classifier-Free Guided Flowscited by 0

50 most cited of 52 All 52 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Conditional Diffusion Models with Classifier-Free Gibbs-like Guidance
  2. 2025Improving Classifier-Free Guidance in Masked Diffusion: Low-Dim Theoretical Insights with High-Dim Impact
  3. 2026P-Guide: Parameter-Efficient Prior Steering for Single-Pass CFG Inference
  4. 2026Classifier-Free Guidance in Flow Matching: Non-Autonomous Potentials, Overshoot, and Posterior-Mean Control
  5. 2025Gaussian Mixture Flow Matching Models
  6. 2025Diffusion Models without Classifier-free Guidance
  7. 2026Momentum Guidance: Plug-and-Play Guidance for Flow Models
  8. 2026Guidance Without Labels: Flow Matching from Coupled Gaussian Mixtures