Spectral Prior for Reducing Exposure Bias in Diffusion Models

Yuya Kobayashi Masato Ishii Yuhta Takida Takashi Shibuya Yuki Mitsufuji

Jul 2026 · arXiv.org · Vol abs/2607.22091 · 0 citations · 34 references

Computer Science

TL;DR

Spectral Alignment is proposed, a lightweight, guidance-based method that calibrates the power spectrum of intermediate predictions to a pre-computed prior and is complementary to Classifier-Free Guidance (CFG).

Abstract

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.

DOI arXiv Semantic Scholar

View source

Similar papers

Preprint Sep 2026

Residual Diffusion Implicit Models

Diffusion models achieve state-of-the-art results across multiple tasks. However, in inverse problems, standard initialization from pure Gaussian noise misaligns the generative process with real-world degradations. More recent methods such as diffusion bridges impose strict endpoint constraints and often require long r...

J. Guerreiro, Pedro Tomás, Helena Aidos et al. · 0 citations

#machine learning Preprint Oct 2026

Error-Corrected Inference-Time Scaling for Imperfect Diffusion Models

Inference-time scaling adapts pretrained diffusion models to new sampling tasks without additional training. Existing methods rely primarily on Monte Carlo sampling with more particles, yet are premised on the pretrained model being exact. In practice, data and training limitations make the model imperfect, and these m...

Zuo-Kai Wen, Louis Grenioux, E. Weinan et al. · 0 citations

Preprint Aug 2026

Revisiting Classifier-Free Guidance Methods in Latent Diffusion Models

This study studies a family of training-free techniques conceptually rooted in Classifier-Free Guidance, most of which were originally proposed on older U-Net diffusion models and validated using metrics that assess image quality in isolation, without accounting for compositional alignment or semantic correspondence.

A. Sergievskii, Artyom Turevich, Sergey Kastryulin · 0 citations

Preprint Sep 2026

ASSERT: Adaptive Stochastic Sampling for Robust Diffusion Models on Analog Compute-in-Memory Hardware

This work investigates diffusion inference using a noise model calibrated and validated against measurements collected from multiple physical CIM chips, and proposes ASSERT, a training-free sampler that uses higher stochasticity early and smoothly transitions to deterministic denoising.

Yuan-Nuo Feng, Yizhe Chen, Wen-Shuai Yao et al. · 0 citations

Preprint Aug 2026

Pixel-Space Diffusion via Observation Operators

Observation Operator Diffusion is proposed, a unified framework that aligns both the supervision trajectory and feature refinement with the intrinsic recovery order of image structures and introduces GL-CoDA, a decoder that injects scale-specific Gaussian-Lanczos observations across decoding stages for coarse-to-fine f...

Shaojie Guo, Li-Chen Ma, Haoyang Tong et al. · 1 citation

Preprint Aug 2026

A Posterior-Dynamics Framework for Imaging Inverse Problems with Pretrained Diffusion Priors

This work defines an explicit surrogate posterior path and derives the Posterior-Dynamics Implicit--Explicit sampler (PD-IMEX), a stable method using one score evaluation per diffusion scale and an implicit data-consistency update, and derives continuous posterior dynamics.

Zhaoqiang Liu, T. Pang, Ruibing Wang et al. · 0 citations