DiffStyle3D: Consistent 3D Gaussian Stylization via Attention Optimization

TL;DR

DiffStyle3D introduces a diffusion-based 3D style transfer method that ensures multi-view consistency by optimizing in latent space with geometry-aware attention mechanisms, leading to improved stylization quality.

Contribution

The paper presents a novel diffusion-based framework with attention and geometry-guided techniques for consistent 3D style transfer, addressing limitations of prior methods.

Findings

Outperforms state-of-the-art methods in stylization quality

Achieves higher multi-view consistency and visual realism

Demonstrates effective geometry-aware optimization in 3D style transfer

Abstract

3D style transfer enables the creation of visually expressive 3D content, enriching the visual appearance of 3D scenes and objects. However, existing VGG- and CLIP-based methods struggle to model multi-view consistency within the model itself, while diffusion-based approaches can capture such consistency but rely on denoising directions, leading to unstable training. To address these limitations, we propose DiffStyle3D, a novel diffusion-based paradigm for 3DGS style transfer that directly optimizes in the latent space. Specifically, we introduce an Attention-Aware Loss that performs style transfer by aligning style features in the self-attention space, while preserving original content through content feature alignment. Inspired by the geometric invariance of 3D stylization, we propose a Geometry-Guided Multi-View Consistency method that integrates geometric information into self-attention to enable cross-view correspondence modeling. Based on geometric information, we additionally construct a geometry-aware mask to prevent redundant optimization in overlapping regions across views, which further improves multi-view consistency. Extensive experiments show that DiffStyle3D outperforms state-of-the-art methods, achieving higher stylization quality and visual realism.