AnchorFlow: Training-Free 3D Editing via Latent Anchor-Aligned Flows

TL;DR

AnchorFlow introduces a training-free 3D editing method that maintains semantic coherence and geometric stability by enforcing latent anchor consistency, enabling robust and pronounced shape modifications without masks.

Contribution

It proposes a novel latent anchor consistency principle with a relaxed alignment loss, improving stability and fidelity in 3D shape editing without model finetuning or masks.

Findings

Outperforms existing methods on Eval3DEdit benchmark

Produces semantically aligned and structurally robust edits

Operates without mask supervision

Abstract

Training-free 3D editing aims to modify 3D shapes based on human instructions without model finetuning. It plays a crucial role in 3D content creation. However, existing approaches often struggle to produce strong or geometrically stable edits, largely due to inconsistent latent anchors introduced by timestep-dependent noise during diffusion sampling. To address these limitations, we introduce AnchorFlow, which is built upon the principle of latent anchor consistency. Specifically, AnchorFlow establishes a global latent anchor shared between the source and target trajectories, and enforces coherence using a relaxed anchor-alignment loss together with an anchor-aligned update rule. This design ensures that transformations remain stable and semantically faithful throughout the editing process. By stabilizing the latent reference space, AnchorFlow enables more pronounced semantic modifications. Moreover, AnchorFlow is mask-free. Without mask supervision, it effectively preserves geometric fidelity. Experiments on the Eval3DEdit benchmark show that AnchorFlow consistently delivers semantically aligned and structurally robust edits across diverse editing types. Code is at https://github.com/ZhenglinZhou/AnchorFlow.