DReSG: Diffusion Residuals for Stylized Gaussian Splatting

Zhongliang Liu1, Wenjie Liu2, Yang Li2,3,†
1School of Software Engineering, East China Normal University, Shanghai, China
2School of Computer Science and Technology, East China Normal University, Shanghai, China
3School of Intelligent Interaction, East China Normal University, Shanghai, China
†Corresponding author
Pacific Graphics 2026
Paper arXiv Code

Teaser

Abstract

Reference-guided stylization of scenes represented by 3D Gaussian Splatting (3DGS) is important for efficient and controllable 3D content creation. Existing VGG-feature-based 3D stylization methods provide stable rendered-view optimization, but often under-represent expressive reference style cues; diffusion models offer stronger image priors, yet direct per-view or score-based diffusion guidance can lead to view drift, local artifacts, and hard-to-control appearance updates. We present DReSG, a 3D-grounded residual-feedback framework for stylized Gaussian splatting. DReSG represents attention-guided diffusion proposals as residual targets relative to the current render, and progressively absorbs these residuals into a shared Gaussian scene through multi-view Gaussian feedback. To make this feedback stable and controllable, DReSG modulates residual strength during target construction and combines coverage-aware view selection with conflict-filtered color updates during multi-view fitting. Extensive experiments demonstrate that DReSG achieves competitive reference-guided stylization while better preserving scene structure and cross-view stability.

Pipeline

DReSG pipeline from attention-guided diffusion proposals to residual targets and multi-view Gaussian feedback.

Qualitative Results

Attention and Residual Strength

3D-Grounded Feedback

Residual Target Construction

User Study

Pairwise user study preferences for style match, content preservation, and overall quality.
DReSG receives more than half of the overall-preference votes against every baseline, ranging from 61.11% to 81.67%.

BibTeX

@misc{liu2026dresg,
  title         = {DReSG: Diffusion Residuals for Stylized Gaussian Splatting},
  author        = {Liu, Zhongliang and Liu, Wenjie and Li, Yang},
  year          = {2026},
  eprint        = {2608.29048},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV},
  url           = {https://arxiv.org/abs/2608.29048},
  note          = {Accepted to Pacific Graphics 2026 (Conference Track)}
}