Awesome-Diffusion-Guidance

Welcome to the collection of research on diffusion guidance methods!
This repo curates cutting-edge papers on diffusion model guidance 🌟

Your stars fuel the updates! ⭐ This library is actively maintained, with new papers and features added regularly.

Table of Contents

• Papers by Topic
  • Foundations & Theory of Diffusion Guidance
  • CFG Correction & High-Guidance-Scale Methods
  • Adaptive, Scheduled & Efficient Guidance
  • Self-, Attention- & Internal-Model Guidance
  • External-Objective & General Conditional Guidance
  • Flow, Alternative Conditioning & Guidance-Free Methods
• How to Contribute
• To Do

Papers by Topic

Tags provide additional descriptors for cross-cutting research properties.

1. Foundations & Theory of Diffusion Guidance

Papers that study the foundations, distributions, and dynamics of diffusion guidance, including classifier guidance and classifier-free guidance.

• Diffusion Models Beat GANs on Image Synthesis
  2021.12 · NeurIPS 2021 · Code
  Tags: Classifier Guidance · Guidance Foundations
  Introduces classifier guidance using gradients to enhance sample quality in diffusion models, surpassing GANs on image synthesis.

• Classifier-Free Diffusion Guidance
  2021.12 · NeurIPS 2021 Workshop
  Tags: Classifier-Free Guidance · CFG Foundations
  Enables guidance without classifiers by jointly training on conditional and unconditional objectives.

• Characteristic Guidance: Non-linear Correction for Diffusion Model at Large Guidance Scale
  2023.12 · ICML 2024 · Code
  Tags: CFG Theory · Fokker–Planck · High Guidance Scale
  Applies non-linear corrections based on Fokker-Plank equations to handle over-saturation artifact at large guidance scales in diffusion models.

• Theoretical Insights for Diffusion Guidance: A Case Study for Gaussian Mixture Models
  2024.03 · ICML 2024
  Tags: CFG Theory · Probabilistic Analysis · Gaussian Mixture Models
  Provide theoretical study towards understanding the influence of guidance on diffusion models in the context of Gaussian mixture models.

• Inner Classifier-Free Guidance and Its Taylor Expansion for Diffusion Models
  2024.04 · TMLR
  Tags: CFG Theory · CFG Alternative · Taylor Expansion
  This approach involves utilizing a single model to jointly optimize the conditional score predictor and unconditional score predictor, eliminating the need for additional classifiers. The inner classifier-free guidance (ICFG) provides an alternative perspective on the CFG method when the condition has a specific structure, demonstrating that CFG represents a first-order case of ICFG.

• REG: Rectified Gradient Guidance for Conditional Diffusion Models
  2025.01 · ICML 2025
  Tags: CFG Theory · CFG Correction · Generalized Guidance
  Proposes a rectified gradient guidance method that enhances conditional diffusion models by aligning practical implementations with a theoretically valid scaled joint distribution objective, improving performance over existing guidance techniques.

• Classifier-Free Guidance: From High-Dimensional Analysis to Generalized Guidance Forms
  2025.02 · arXiv
  Tags: CFG Theory · Probabilistic Analysis · Generalized Guidance
  Uniquely demonstrates that Classifier-Free Guidance accurately reproduces the target distribution in high and infinite dimensions, extending to a family of non-linear generalizations with improved robustness, sample fidelity, and diversity.

• Conditional Diffusion Models with Classifier-Free Gibbs-like Guidance
  2025.05 · arXiv · Code
  Tags: CFG Theory · Probabilistic Analysis · Diversity
  Samples tilted distributions with Rényi divergence for low-noise corrections, fixing CFG's diversity issues.

2. CFG Correction & High-Guidance-Scale Methods

Papers that correct, stabilize, or reinterpret classifier-free guidance, especially at high guidance scales.

• Characteristic Guidance: Non-linear Correction for Diffusion Model at Large Guidance Scale
  2023.12 · ICML 2024 · Code
  Tags: CFG Theory · Fokker–Planck · High Guidance Scale
  Applies non-linear corrections based on Fokker-Plank equations to handle over-saturation artifact at large guidance scales in diffusion models.

• CFG++: Manifold-constrained Classifier Free Guidance for Diffusion Models
  2024.06 · ICLR 2025 · Code
  Tags: CFG Correction · Manifold · High Guidance Scale
  Constrains guidance to data manifold using orthogonal projections for better sample quality and invertibility.

• Eliminating Oversaturation and Artifacts of High Guidance Scales in Diffusion Models
  2024.10 · ICLR 2025 · Code
  Tags: CFG Correction · Geometry · High Guidance Scale
  Decompose the update term in CFG into parallel and orthogonal components with respect to the conditional model prediction and observe that the parallel component primarily causes oversaturation, while the orthogonal component enhances image quality.

• TCFG: Tangential Damping Classifier-free Guidance
  2025.03 · CVPR 2025 · Code
  Tags: CFG Correction · Geometry · High Guidance Scale
  Geometrically filters tangential misalignments using SVD for better manifold alignment and quality with minimal overhead.

• Angle Domain Guidance: Latent Diffusion Requires Rotation Rather Than Extrapolation
  2025.06 · ICML 2025 · Code
  Tags: CFG Correction · Geometry · High Guidance Scale
  Angle Domain Guidance (ADG) mitigates color distortions in text-to-image latent diffusion models by constraining magnitude variations and optimizing angular alignment, preserving enhanced text-image alignment at higher guidance weights.

• Guidance in the Frequency Domain Enables High-Fidelity Sampling at Low CFG Scales
  2025.06 · ICLR 2026 · Code
  Tags: CFG Correction · Frequency Domain · High Guidance Scale
  Analyzes CFG in the frequency domain and proposes frequency-decoupled guidance (FDG): separate scales on low- vs. high-frequency parts of the CFG update (structure and conditioning vs. detail), improving fidelity at low guidance while reducing high-CFG oversaturation and diversity loss.

• CFG-Ctrl: Control-Based Classifier-Free Diffusion Guidance
  2026.03 · CVPR 2026 · Code
  Tags: CFG Correction · Control Theory · High Guidance Scale
  Reframes CFG as control on the generative flow and proposes SMC-CFG (sliding mode control) with nonlinear feedback to reduce instability and overshooting at large guidance scales; Code release includes FLUX, Qwen-Image, SD3/SD3.5, and Wan video pipelines.

3. Adaptive, Scheduled & Efficient Guidance

Papers that adapt guidance strength, schedule guidance over time, or reduce the computational cost of guided sampling.

• Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models
  2023.12 · AAAI 2025
  Tags: Adaptive Guidance · Guidance Schedule · Efficiency
  Skips evaluations upon convergence via NAS policies for training-free acceleration reducing NFEs.

• Analysis of Classifier-Free Guidance Weight Schedulers
  2024.04 · TMLR
  Tags: Adaptive CFG · Guidance Schedule · CFG Analysis
  Investigates varying classifier-free guidance weights during diffusion, finding that simple, monotonically increasing schedulers improve performance with minimal code, while complex parametrized schedulers offer gains but lack generalizability across models and tasks.

• Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models
  2024.04 · NeurIPS 2024 · Code
  Tags: Guidance Schedule · Efficiency · CFG
  Restricts classifier-free guidance to a limited noise level interval during sampling, improving inference speed and result quality.

• Classifier-free Guidance with Adaptive Scaling
  2025.02 · arXiv
  Tags: Adaptive CFG · Dynamic Guidance · Guidance Schedule
  β-CFG introduces a novel adaptive scaling method using gradient-based normalization and time-dependent β-distribution curves to dynamically balance prompt matching and image quality during the diffusion denoising process.

• Adaptive Diffusion Guidance via Stochastic Optimal Control
  2025.05 · arXiv
  Tags: Adaptive Guidance · Optimal Control · Guidance Schedule
  Introduces a stochastic optimal control framework that dynamically adjusts guidance strength in diffusion models based on time, current sample, and conditioning class, offering a principled approach to guidance scheduling.

• Feedback Guidance of Diffusion Models
  2025.06 · arXiv · Code
  Tags: Feedback · Adaptive Guidance · Dynamic Guidance
  Self-regulates coefficients based on predicted informativeness, adapting for complex prompts and balancing diversity-quality.

• How Much To Guide: Revisiting Adaptive Guidance in Classifier-Free Guidance Text-to-Vision Diffusion Models
  2025.06 · arXiv
  Tags: Adaptive CFG · Guidance Schedule · Efficiency
  Step AG, a simple and universally applicable adaptive guidance strategy, restricts classifier-free guidance to the first several denoising steps, achieving high-quality, well-conditioned images with a 20% to 30% speedup.

• Dynamic Classifier-Free Diffusion Guidance via Online Feedback
  2025.09 · arXiv
  Tags: Feedback · Adaptive CFG · Guidance Schedule
  Introduces a dynamic CFG framework using online feedback from latent-space evaluators (e.g., CLIP, discriminators) to adaptively adjust guidance scales per timestep, achieving up to 55.5% human preference win-rate for prompt-specific image generation.

• Towards a Golden Classifier-Free Guidance Path via Foresight Fixed Point Iterations
  2025.10 · NeurIPS 2025 · Code
  Tags: Fixed-Point · CFG Theory · Efficiency
  Introduces Foresight Guidance (FSG), a unified framework that reframes CFG and its variants as fixed point iterations seeking a golden path. FSG prioritizes longer-interval subproblems in early diffusion stages with increased iterations, achieving state-of-the-art performance in both image quality and computational efficiency.

4. Self-, Attention- & Internal-Model Guidance

Papers that derive guidance signals from self-attention, internal representations, or related signals within the diffusion model.

• Attend-and-Excite: Attention-Based Semantic Guidance for Text-to-Image Diffusion Models
  2023.01 · ACM Transactions on Graphics
  Tags: Cross-Attention · Semantic Guidance · Training-Free
  Introduces Attend-and-Excite, an attention-based semantic guidance method that iteratively refines cross-attention maps to ensure text-to-image diffusion models faithfully generate all subjects in the prompt (requires new attention processor per model).

• Diffusion Self-Guidance for Controllable Image Generation
  2023.06 · NeurIPS 2023
  Tags: Self-Guidance · Internal Signal · Controllable Generation
  Introduces self-guidance, a method that provides greater control over generated images by guiding the internal representations of diffusion models.

• Improving Sample Quality of Diffusion Models Using Self-Attention Guidance
  2023.10 · ICCV 2023 · Code
  Tags: Self-Attention · Training-Free · Artifact Reduction
  Uses self-attention maps to guide diffusion models away from degraded regions, improving generated image quality without extra training.

• Self-Rectifying Diffusion Sampling with Perturbed-Attention Guidance
  2024.03 · ECCV 2024 · Code
  Tags: PAG · Attention Guidance · Training-Free
  Proposes Perturbed-Attention Guidance (PAG) to improve sample quality in diffusion models by adding random perturbations to attention maps, enabling self-rectification in both unconditional and conditional generation.

• Guiding a Diffusion Model with a Bad Version of Itself
  2024.06 · NeurIPS 2024 · Code
  Tags: Weak-to-Strong · Internal Signal · Self-Guidance
  Uses a poorly trained model version for guidance to control trade-offs between sample quality and diversity.

• Smoothed Energy Guidance: Guiding Diffusion Models with Reduced Energy Curvature of Attention
  2024.08 · NeurIPS 2024 · Code
  Tags: Attention Guidance · Energy Guidance · Artifact Reduction
  Presents Smoothed Energy Guidance (SEG), a method that smooths energy curvature in attention maps to enhance unconditional image generation quality while minimizing artifacts (requires new attention processor per model).

• Spatiotemporal Skip Guidance for Enhanced Video Diffusion Sampling
  2024.11 · CVPR 2025 · Code
  Tags: Skip Guidance · Internal Signal · Video Diffusion
  Develops Spatiotemporal Skip Guidance (STG), a training-free technique that skips specific layers in video diffusion transformers to boost sample quality without sacrificing diversity or motion dynamics.

• Entropy Rectifying Guidance for Diffusion and Flow Models
  2025.04 · arXiv
  Tags: Attention Guidance · Flow Guidance · Training-Free
  Entropy Rectifying Guidance (ERG) is a simple and effective guidance mechanism that improves image quality, diversity, and prompt consistency in diffusion and flow models by modifying the attention mechanism during inference, extending to unconditional sampling and combining seamlessly with other guidance methods.

• Normalized Attention Guidance: Universal Negative Guidance for Diffusion Model
  2025.05 · NeurIPS 2025
  Tags: Attention Guidance · Negative Guidance · Training-Free
  Introduces a training-free, universal negative guidance method for diffusion models that uses extrapolation in attention space with L1-based normalization, generalizing across architectures, sampling regimes, and modalities while maintaining fidelity.

• Token Perturbation Guidance for Diffusion Models
  2025.06 · NeurIPS 2025 · Code
  Tags: Token Perturbation · Internal Signal · Training-Free
  Shuffles tokens for training-free, condition-agnostic guidance improving unconditional generation.

• S²-Guidance: Stochastic Self Guidance for Training-Free Enhancement of Diffusion Models
  2025.08 · arXiv · Code
  Tags: Self-Guidance · Skip Guidance · Training-Free
  Introduces a training-free self-guidance method using stochastic block-dropping to create sub-networks that refine suboptimal predictions, enhancing sample quality and prompt adherence beyond traditional CFG.

• Improving Diffusion Generalization with Weak-to-Strong Segmented Guidance
  2026.03 · CVPR 2026 · Code
  Tags: Weak-to-Strong · Adaptive Guidance · Training-Free
  This paper introduces the Weak‑to‑Strong (W2S) guidance principle and Segmented Guidance (SGG), a framework that combines condition‑dependent and condition‑agnostic guidance to improve the generalization of diffusion models.

5. External-Objective & General Conditional Guidance

Papers that use external objectives, discriminators, energy functions, or other general conditioning signals to guide generation.

• Refining Generative Process with Discriminator Guidance in Score-based Diffusion Models
  2022.11 · ICML 2023 · Code
  Tags: Discriminator Guidance · External Objective
  Introduces Discriminator Guidance, which improves sample quality in pre-trained diffusion models by separately training a discriminator to assess denoising path realism and adding an auxiliary term during generation to deceive it, correcting the score to approximate the data distribution without GAN-like joint training.

• Training-Free Structured Diffusion Guidance for Compositional Text-to-Image Synthesis
  2022.12 · ICLR 2023 · Code
  Tags: Training-Free · Structured Guidance · Semantic Guidance
  Improve the compositional skills of T2I models, specifically more accurate attribute binding and better image compositions, by incorporating linguistic structures with the diffusion guidance process.

• Universal Guidance for Diffusion Models
  2023.06 · CVPR Workshops 2023 · Code
  Tags: Training-Free · Energy Guidance · Arbitrary Conditions
  Provides a training-free method to incorporate arbitrary conditions into pre-trained unguided diffusion models using energy functions.

• TFG: Unified Training-Free Guidance for Diffusion Models
  2024.09 · NeurIPS 2024 · Code
  Tags: Training-Free · External Objective · General Conditional Guidance
  Unifies training-free guidance methods with efficient hyper-parameter optimization for various conditional generation tasks.

• Variational Control for Guidance in Diffusion Models
  2025.02 · arXiv · Code
  Tags: Training-Free · External Objective · Inverse Problems
  Employs variational inference for terminal cost guidance in pre-trained models, unifying methods for training-free inverse problems.

• Improving Discriminator Guidance in Diffusion Models
  2025.03 · arXiv
  Tags: Discriminator Guidance · External Objective · Probabilistic Analysis
  Proposes a training objective for discriminator guidance in diffusion models that minimizes Kullback-Leibler divergence, improving sample quality over the conventional method using Cross-Entropy loss.

• Domain Guidance: A Simple Transfer Approach for a Pre-trained Diffusion Model
  2025.04 · ICLR 2025
  Tags: Domain Transfer · General Conditional Guidance · Training-Free
  Introduces a simple transfer approach that leverages pre-trained knowledge to guide the sampling process toward the target domain, sharing a formulation similar to classifier-free guidance for improved domain alignment and generation quality.

• Instructing Text-to-Image Diffusion Models via Classifier-Guided Semantic Optimization
  2025.05 · arXiv
  Tags: External Objective · Semantic Guidance · Training-Free
  Proposes optimizing semantic embeddings guided by attribute classifiers to steer text-to-image diffusion models for desired edits, eliminating the need for text prompts and model training or fine-tuning.

• LumaGuide: Distribution Shaping for Training-Free HDR Generation in Diffusion Models
  2026.08 · arXiv 2026 · Code
  Tags: Training-Free · Energy Guidance · Distribution Shaping
  Training-free, energy-based guidance that steers sampling toward a target feature distribution instead of a class or prompt. Instantiated for HDR generation by matching a luminance histogram in PQ space through the VAE with a Wasserstein-1 loss on Flux.1-dev; targets come from a small text-to-histogram regressor, a reference image, or a preset, and the same machinery extends to video with temporal consistency terms.

6. Flow, Alternative Conditioning & Guidance-Free Methods

Papers that explore flow matching, alternative conditioning mechanisms, or generation without conventional guidance.

• ProtoDiffusion: Classifier-Free Diffusion Guidance with Prototype Learning
  2024.02 · ACML 2024 · Code
  Tags: Prototype Guidance · CFG Alternative · Efficiency
  Integrates prototype learning into classifier-free guidance to speed up sampling and enhance quality using class-specific prototypes.

• No Training, No Problem: Rethinking Classifier-Free Guidance for Diffusion Models
  2024.07 · ICLR 2025
  Tags: CFG Alternative · CFG Theory · Training-Free
  Introduces independent condition guidance (ICG) and time-step guidance (TSG) using time-step encodings for training-free, invertible sampling applicable unconditionally.

• TFG-Flow: Training-free Guidance in Multimodal Generative Flow
  2025.01 · ICLR 2025
  Tags: Flow Matching · Training-Free · Flow Guidance
  Introduces a novel training-free guidance method for multimodal generative flow, uniquely addressing the curse-of-dimensionality while maintaining unbiased sampling for guiding discrete variables in applications like molecular design.

• Visual Generation Without Guidance
  2025.01 · ICML 2025 · Code
  Tags: Guidance-Free · CFG Alternative · Conditional Generation
  Uses only a single model by reformulating the training loss as a linear interpolation between conditional and unconditional components with a randomly sampled pseudo-temperature and a stopping gradient for stability, halving inference costs while matching quality and diversity.

• Diffusion Models without Classifier-free Guidance
  2025.02 · arXiv · Code
  Tags: CFG-Free · Explicit Conditioning · Inference Efficiency
  Integrates condition posteriors into objectives for accelerated processes matching CFG quality without unconditional models.

• CFG-Zero*: Improved Classifier-Free Guidance for Flow Matching Models
  2025.03 · arXiv · Code
  Tags: Flow Matching · CFG Correction · Guidance Schedule
  Optimizes guidance scales for flow matching to correct early velocities and zero initial steps, outperforming CFG in multimedia generation.

• Guidance Free Image Editing via Explicit Conditioning
  2025.03 · arXiv
  Tags: Guidance-Free · Explicit Conditioning · Image Editing
  Explicit Proposes a technique that models noise distribution on input modalities to guide conditional diffusion models, significantly reducing computational costs and improving inference time compared to CFG in image editing tasks.

• Diffusion Models with Double Guidance: Generate with aggregated datasets
  2025.05 · arXiv
  Tags: Multi-Condition · CFG Alternative · Conditional Generation
  Enables precise conditional generation by maintaining control over multiple conditions without requiring joint annotations, even when training samples lack all conditions simultaneously.

• Rectified-CFG++ for Flow Based Models
  2025.10 · NeurIPS 2025 · Code
  Tags: Rectified Flow · Flow Guidance · CFG Correction
  An adaptive predictor-corrector guidance method for rectified flow models that prevents off-manifold drift, ensuring stable, high-quality, and artifact-free image generation across a wide range of guidance scales .

How to Contribute

If you have a relevant paper to add, please follow these steps:

1. Fork the repository.
2. Create a new branch for your changes.
3. Add the paper to the appropriate topic section.
4. Include the paper link, date, venue, tags, code link when available, and a concise neutral summary.
5. Commit and push your changes.
6. Open a pull request for review.

To Do

• Continue adding relevant papers.
• Periodically verify paper, venue, and code links.