Spectral Prior for Reducing Exposure Bias in Diffusion Models

Hugging Face Daily Papers Papers
diffusion-models exposure-bias spectral-prior guidance image-generation

Summary

This paper proposes Spectral Alignment (SPA), a lightweight guidance-based method that reduces exposure bias in diffusion models by calibrating the power spectrum of intermediate predictions, showing consistent improvements across pixel-space, latent, and flow-matching models with minimal computational overhead.

Diffusion models typically suffer from error accumulation during iterative sampling, commonly referred to as exposure bias. We reveal systematic frequency-dependent discrepancies between training and inference, which can be interpreted as frequency-dependent SNR error. Crucially, the direction of this mismatch varies across models and timesteps, indicating that fixed correction rules do not generalize. We propose Spectral Alignment (SPA), a lightweight, guidance-based method that calibrates the power spectrum of intermediate predictions to a pre-computed prior. Our approach consists of two stages: (1) offline fitting of a parametric spectrum model from training data, and (2) inference-time guidance via efficient FFT-based gradient computation. SPA introduces minimal computational overhead (3-4\%) and is complementary to Classifier-Free Guidance (CFG). We demonstrate consistent improvements across diverse architectures, from pixel-space models (DDPM, ADM) to latent diffusion models (SD2.0, SDXL) and flow-matching models (SD3.5, FLUX). Our implementation is available at https://github.com/SonyResearch/SPA.
Original Article
View Cached Full Text

Cached at: 07/27/26, 09:43 AM

Paper page - Spectral Prior for Reducing Exposure Bias in Diffusion Models

Source: https://huggingface.co/papers/2607.22091

Abstract

Diffusionmodelstypicallysufferfromerroraccumulationduringiterativesampling,commonlyreferredtoasexposurebias.Werevealsystematicfrequency-dependentdiscrepanciesbetweentrainingandinference,whichcanbeinterpretedasfrequency-dependentSNRerror.Crucially,thedirectionofthismismatchvariesacrossmodelsandtimesteps,indicatingthatfixedcorrectionrulesdonotgeneralize.WeproposeSpectralAlignment(SPA),alightweight,guidance-basedmethodthatcalibratesthepowerspectrumofintermediatepredictionstoapre-computedprior.Ourapproachconsistsoftwostages:(1)offlinefittingofaparametricspectrummodelfromtrainingdata,and(2)inference-timeguidanceviaefficientFFT-basedgradientcomputation.SPAintroducesminimalcomputationaloverhead(3-4\%)andiscomplementarytoClassifier-FreeGuidance(CFG).Wedemonstrateconsistentimprovementsacrossdiversearchitectures,frompixel-spacemodels(DDPM,ADM)tolatentdiffusionmodels(SD2.0,SDXL)andflow-matchingmodels(SD3.5,FLUX).Ourimplementationisavailableathttps://github.com/SonyResearch/SPA.

View arXiv pageView PDFGitHub0Add to collection

Get this paper in your agent:

hf papers read 2607\.22091

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.22091 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.22091 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.22091 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Spectral Feedback for Test-Time Alignment of Protein Diffusion Models

arXiv cs.AI

The paper presents Spectral Feedback, an algorithm that enhances test-time alignment for discrete diffusion models in protein inverse folding by iteratively selecting edit-positions using sparse Fourier representations, resulting in improved performance for reward maximization.

Elucidating the SNR-t Bias of Diffusion Probabilistic Models

Hugging Face Daily Papers

This paper identifies a Signal-to-Noise Ratio timestep (SNR-t) bias in diffusion probabilistic models during inference, where SNR-timestep alignment from training is disrupted at inference time. The authors propose a differential correction method that decomposes samples into frequency components and corrects each separately, improving generation quality across models like IDDPM, ADM, DDIM, EDM, and FLUX with minimal computational overhead.

Learning to Discretize: Diffusion-Based Adaptive Mesh with Spectral Guidance

arXiv cs.LG

This paper proposes a diffusion-based framework for learning adaptive mesh discretization conditioned on observed PDE dynamics, using spectral guidance and physics constraints to allocate resolution where needed. The method achieves competitive or superior performance across five PDE regimes.

Specificity-Aware Diffusion Steering via Variance-Reduced Sequential Monte Carlo

arXiv cs.LG

The paper introduces a specificity-aware diffusion steering method that designs an explicit target distribution from an overlap-based objective and samples it with a variance-reduced Sequential Monte Carlo sampler, improving negative-guidance effectiveness and sampling stability across image generation and peptide-MHC binder tasks.