Style Injection in Diffusion: A Training-free Approach for Adapting Large-scale Diffusion Models for Style Transfer

TL;DR

This paper presents a training-free style transfer method using large-scale diffusion models by manipulating self-attention features, enabling efficient and high-quality artistic style transfer without optimization.

Contribution

The authors introduce a novel approach that modifies self-attention in diffusion models for style transfer, eliminating the need for inference-time optimization.

Findings

Outperforms state-of-the-art style transfer methods

Preserves content while transferring style effectively

Handles color and content disruptions with proposed techniques

Abstract

Despite the impressive generative capabilities of diffusion models, existing diffusion model-based style transfer methods require inference-stage optimization (e.g. fine-tuning or textual inversion of style) which is time-consuming, or fails to leverage the generative ability of large-scale diffusion models. To address these issues, we introduce a novel artistic style transfer method based on a pre-trained large-scale diffusion model without any optimization. Specifically, we manipulate the features of self-attention layers as the way the cross-attention mechanism works; in the generation process, substituting the key and value of content with those of style image. This approach provides several desirable characteristics for style transfer including 1) preservation of content by transferring similar styles into similar image patches and 2) transfer of style based on similarity of local texture (e.g. edge) between content and style images. Furthermore, we introduce query preservation and attention temperature scaling to mitigate the issue of disruption of original content, and initial latent Adaptive Instance Normalization (AdaIN) to deal with the disharmonious color (failure to transfer the colors of style). Our experimental results demonstrate that our proposed method surpasses state-of-the-art methods in both conventional and diffusion-based style transfer baselines.