Where and How to Perturb: On the Design of Perturbation Guidance in Diffusion and Flow Models

TL;DR

This paper introduces HeadHunter and SoftPAG, novel methods for attention head perturbation in diffusion models, enabling fine-grained, interpretable control over image generation quality and style attributes.

Contribution

It provides the first head-level analysis of attention perturbation, revealing head specialization and proposing systematic, user-centric perturbation strategies for diffusion models.

Findings

Attention heads govern distinct visual concepts such as structure, style, and texture.

HeadHunter effectively selects attention heads aligned with user objectives.

SoftPAG offers continuous control over perturbation strength, improving generation quality.

Abstract

Recent guidance methods in diffusion models steer reverse sampling by perturbing the model to construct an implicit weak model and guide generation away from it. Among these approaches, attention perturbation has demonstrated strong empirical performance in unconditional scenarios where classifier-free guidance is not applicable. However, existing attention perturbation methods lack principled approaches for determining where perturbations should be applied, particularly in Diffusion Transformer (DiT) architectures where quality-relevant computations are distributed across layers. In this paper, we investigate the granularity of attention perturbations, ranging from the layer level down to individual attention heads, and discover that specific heads govern distinct visual concepts such as structure, style, and texture quality. Building on this insight, we propose "HeadHunter", a systematic framework for iteratively selecting attention heads that align with user-centric objectives, enabling fine-grained control over generation quality and visual attributes. In addition, we introduce SoftPAG, which linearly interpolates each selected head's attention map toward an identity matrix, providing a continuous knob to tune perturbation strength and suppress artifacts. Our approach not only mitigates the oversmoothing issues of existing layer-level perturbation but also enables targeted manipulation of specific visual styles through compositional head selection. We validate our method on modern large-scale DiT-based text-to-image models including Stable Diffusion 3 and FLUX.1, demonstrating superior performance in both general quality enhancement and style-specific guidance. Our work provides the first head-level analysis of attention perturbation in diffusion models, uncovering interpretable specialization within attention layers and enabling practical design of effective perturbation strategies.