Easy3E: Feed-Forward 3D Asset Editing via Rectified Voxel Flow

Shimin Hu; Yuanyi Wei; Fei Zha; Yudong Guo; Juyong Zhang

cs.CVFeb 25, 2026

Easy3E: Feed-Forward 3D Asset Editing via Rectified Voxel Flow

Authors: Shimin Hu, Yuanyi Wei, Fei Zha, Yudong Guo, Juyong Zhang

Organizations: University of Science and Technology of China

Abstract

Existing 3D editing methods rely on computationally intensive scene-by-scene iterative optimization and suffer from multi-view inconsistency. We propose an effective and feed-forward 3D editing framework based on the TRELLIS generative backbone, capable of modifying 3D models from a single editing view. Our framework addresses two key issues: adapting training-free 2D editing to structured 3D representations, and overcoming the bottleneck of appearance fidelity in compressed 3D features. To ensure geometric consistency, we introduce Voxel FlowEdit, an edit-driven flow in the sparse voxel latent space that achieves globally consistent 3D deformation in a single pass. To restore high-fidelity details, we develop a normal-guided single to multi-view generation module as an external appearance prior, successfully recovering high-frequency textures. Experiments demonstrate that our method enables fast, globally consistent, and high-fidelity 3D model editing.

Figures & tables

Figure 2: Overview of Easy3E. The framework operates in two main stages: Geometry Editing and Texture Refinement. Starting from a rendered source view, an edited target image provides the guidance for editing. In the Geometry Editing stage, the Voxel FlowEdit algorithm transforms the source voxel structure under flow-based guidance, followed by SLAT Repainting that refines local latent features to produce the target mesh. The Texture Refinement stage then employs a generation branch and a normal-guided control adapter to synthesize multi-view-consistent textures, which are projected and fused onto the mesh to yield the final high-fidelity 3D asset.

Figure 3: Comparison of FlowEdit’s limitations and the Voxel FlowEdit solution. (a) Base FlowEdit: The semantic velocity $\mathbf{v}_{\text{edit}}$ is corrupted by accumulated approximation error, causing the trajectory to drift and resulting in structural corruption (red dashed box). (b) Voxel FlowEdit: The edit is driven by external gradient guidance $\mathbf{G}_{\text{sil}}$ , while internal correction $\boldsymbol{\xi}_{\text{traj}}$ maintains manifold consistency. This combined approach achieves a clean and structurally integral edit.

Figure 4: Qualitative comparison. Our method achieves clean geometry and consistent appearance across multiple views, faithfully realizing the target edits while strictly preserving unedited regions. Competing methods either retain the original geometry (MVEdit) or exhibit strong structural distortion and inconsistency (Vox-E, Instant3dit). Although VoxHammer also utilizes mask inputs to guide the editing process, it struggles to maintain high fidelity outside the masked areas.

Method	CLIP-T $\uparrow$	DINO-I $\uparrow$	LPIPS $\downarrow$	FID $\downarrow$
TRELLIS [ 48 ]	0.323	0.895	0.243	30.2
MVEdit [ 6 ]	0.267	0.851	0.282	67.6
Vox-E [ 40 ]	0.266	0.734	0.673	90.3
Instant3DiT [ 1 ]	0.285	0.874	0.286	49.7
VoxHammer [ 21 ]	0.295	0.938	0.193	31.7
Ours	0.326	0.952	0.138	25.8

Table 1: Quantitative comparison on the 3D editing benchmark. Higher is better for CLIP-T, and DINO-I; lower is better for LPIPS and FID. Our method consistently achieves the best performance across all metrics.

Question	Q1 $\uparrow$	Q2 $\uparrow$	Q3 $\uparrow$	Q4 $\uparrow$	Q5 $\uparrow$
Ours	88.98	94.63	94.92	97.51	97.00

Table 2: User study results. Percentage of times each question is rated best (higher is better).

Figure 5: Comparison between without and with Flow Guidance. Disabling $\mathbf{G}_{\text{sil}}$ and $\boldsymbol{\xi}_{\text{traj}}$ jointly leads to structural collapse and view-inconsistent deformation, whereas enabling both yields stable and silhouette-aligned geometry.

Figure 6: Comparison between without and with Texture Refinement. The refinement stage significantly enhances surface detail and view-consistent appearance.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Method	Runtime
Vox-E [ 40 ]	37 min
MVEdit [ 6 ]	212 s
Instant3dit [ 1 ]	25 s
Ours	75 s

Appendix

Table 3: Runtime comparison with different methods.

Figure 7: Visualizations of the normal-guided multi-view diffusion module. Taking rendered normal maps and a reference image as input, the module generates multi-view images that are both texture-rich and geometrically consistent, serving as robust priors for 3D texture recovery.

3D Asset Editing

See all 18 papers

Feedforward 3D Editing Learns from Semantic-Part Transformation
May 26, 2026Jiawei Weng, Saining Zhang, Zhenxin Diao +43D Asset Generation 3D Asset Editing
Tsinghua University, China
Omni-3DEdit: Generalized Versatile 3D Editing in One-Pass
Mar 18, 2026Chen Liyi, Wang Pengfei, Zhang Guowen +23D Scene Editing 3D Shape Editing
The Hong Kong Polytechnic University
EditFlow3D: Automated Local Editing of 3D Assets with Trajectory Preservation
Aug 4, 2026Rui Nie, Chuang Wang, Haitao Zhou +43D Asset Editing 3D Shape Editing
Beihang University · Bambu Lab · Shanghai Jiao Tong University

May 26, 2026J/K move · Enter open · S save

Feedforward 3D Editing Learns from Semantic-Part Transformation

Jiawei Weng, Saining Zhang, Zhenxin Diao, Peishuo Li+3

Tsinghua University, China

3D Asset Generation 3D Asset Editing 3D Shape Editing

Open paper AskPDF arXiv

3D editing is a fundamental capability for scalable 3D content creation. While image editing has rapidly evolved toward large-scale feedforward generative paradigms, 3D AI generation remains dominated by training-free editing pipelines. A central challenge of feedforward 3D editing lies in the lack of high-quality paired supervision. Editable 3D assets require simultaneous preservation of geometry, multi-view consistency, structural coherence, and localized edit controllability. Existing 3D editing datasets often rely on independently generated assets, image-mediated reconstruction or narrow edit taxonomies, leading to inaccurate localization, weak preservation, blurred edit boundaries, and limited semantic consistency. In this work, we introduce a new perspective: scalable feedforward 3D editing should be learned from semantic-part transformations. Based on this insight, we propose Pxform, a high-quality 3D editing dataset with over 100K consistent before/after editing pairs across seven edit types. Instead of treating objects as unstructured shapes, our pipeline grounds edits directly in semantic 3D parts. Built upon Pxform, we further propose PartFlow, a feedforward 3D editing network that injects source-aware latent control into pretrained 3D generative priors. PartFlow introduces mask-aware velocity preservation and render-space consistency supervision to jointly improve edit fidelity and source preservation, while requiring no 3D edit mask during inference. Extensive experiments demonstrate that high-quality semantic-part supervision substantially improves scalable 3D editing, enabling PartFlow to achieve state-of-the-art performance on both geometric and appearance editing benchmarks.

Abstract

3D Asset Editing

Explore similar work

Feedforward 3D Editing Learns from Semantic-Part Transformation

Omni-3DEdit: Generalized Versatile 3D Editing in One-Pass

EditFlow3D: Automated Local Editing of 3D Assets with Trajectory Preservation