mlsanityData to 14 Oct 2026

Adversarial Learning of Classifier-Free Guidance Schedules

arXiv 2608.14038 ↗First posted 14 Aug 2026cs.LGcited by 0

On the map: Diffusion models → Diffusion foundations and sampling

Abstract (arXiv)

Modern text-to-image diffusion models rely on classifier-free guidance (CFG) to achieve high image fidelity and text alignment. However, CFG typically applies a static, global scale across all timesteps, samples, and conditions -- a choice that is generally suboptimal and can introduce artifacts, as different states may benefit from different levels of guidance. While time-varying schedules are known to improve quality, designing them by hand is non-trivial and application-dependent. In this paper, we learn the guidance schedule as a function of diffusion time, conditioning and the current noisy sample, in order to better align sampled images with the text prompt. We frame this as a density ratio estimation problem: a discriminator is trained to estimate the time-dependent log-density ratio between the true and guided marginal distributions, while a lightweight generator network predicts the optimal, state-dependent guidance scale. Empirically, our approach outperforms both heuristic CFG schedules and prior methods for learning dynamic guidance on text-to-image generation benchmarks.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Diffusion foundations and sampling, one of 14 directions in Diffusion models. Builds on 44 of its 50 references that are in the map. First posted on arXiv 14 Aug 2026.

Builds on

44 of 50 references.

Closest ancestors · 29

Same direction, newest first

  1. 2025Learn to Guide Your Diffusion Model1 Oct
  2. 2025Feedback Guidance of Diffusion Models6 Jun
  3. 2025Provable Efficiency of Guidance in Diffusion Models for General Data Distribution2 May
  4. 2025Improving Discriminator Guidance in Diffusion Models20 Mar
  5. 2025One-step Diffusion Models with f-Divergence Distribution Matching21 Feb
  6. 2025Classifier-free Guidance with Adaptive Scaling14 Feb
  7. 2024Rectified Diffusion Guidance for Conditional Generation24 Oct
  8. 2024Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models3 Oct
  9. 2024What does guidance do? A fine-grained analysis in a simple setting19 Sep
  10. 2024GeoGuide: Geometric guidance of diffusion models17 Jul
  11. 2024No Training, No Problem: Rethinking Classifier-Free Guidance for Diffusion Models2 Jul
  12. 2024CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models12 Jun
17 more
  1. 2024Guiding a Diffusion Model with a Bad Version of Itself4 Jun
  2. 2024Improved Distribution Matching Distillation for Fast Image Synthesis23 May
  3. 2024Analysis of Classifier-Free Guidance Weight Schedulers19 Apr
  4. 2024Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models11 Apr
  5. 2024Understanding and Improving Training-free Loss-based Diffusion Guidance19 Mar
  6. 2024Fast High-Resolution Image Synthesis with Latent Adversarial Diffusion Distillation18 Mar
  7. 2023Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models19 Dec
  8. 2023One-step Diffusion with Distribution Matching Distillation30 Nov
  9. 2023FreeDoM: Training-Free Energy-Guided Conditional Diffusion Model17 Mar
  10. 2023Consistency Models2 Mar
  11. 2023Universal Guidance for Diffusion Models14 Feb
  12. 2022Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models28 Nov
  13. 2022Classifier-Free Diffusion Guidance26 Jul
  14. 2021High-Resolution Image Synthesis with Latent Diffusion Models20 Dec
  15. 2021Tackling the Generative Learning Trilemma with Denoising Diffusion GANs15 Dec
  16. 2020Score-Based Generative Modeling through Stochastic Differential Equations26 Nov
  17. 2020Denoising Diffusion Implicit Models6 Oct

Other references · 15

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,744
  2. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  3. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,355
  4. 2022Flow Matching for Generative Modelingcited by 5,702
  5. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,101
  6. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,424
  7. 2015Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networkscited by 2,054
  8. 2018Spectral Normalization for Generative Adversarial Networkscited by 1,636
  9. 2023Pick-a-Pic: An Open Dataset of User Preferences for Text-to-Image Generationcited by 977
  10. 2023Human Preference Score v2: A Solid Benchmark for Evaluating Human Preferences of Text-to-Image Synthesiscited by 890
  11. 2018Which Training Methods for GANs do actually Converge?cited by 679
  12. 2023Rethinking FID: Towards a Better Evaluation Metric for Image Generationcited by 310
3 more
  1. 2017GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibriumcited by 213
  2. 2022Diffusion-GAN: Training GANs with Diffusioncited by 205
  3. 2025Is CLIP ideal? No. Can we fix it? Yes!cited by 21

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Dynamic Classifier-Free Diffusion Guidance via Online Feedback
  2. 2025How Much To Guide: Revisiting Adaptive Guidance in Classifier-Free Guidance Text-to-Vision Diffusion Models
  3. 2026Information-Theoretic Classifier-Free Guidance with Adaptive Schedule Optimization
  4. 2025Navigating with Annealing Guidance Scale in Diffusion Space
  5. 2025Adjusting Initial Noise to Mitigate Memorization in Text-to-Image Diffusion Models
  6. 2026Guidance Matters: Rethinking the Evaluation Pitfall for Text-to-Image Generation
  7. 2026Revisiting Classifier-Free Guidance Methods in Latent Diffusion Models
  8. 2024Rethinking the Spatial Inconsistency in Classifier-Free Diffusion Guidance