AnchorFlow: Training-Free 3D Editing via Latent Anchor-Aligned Flows
This addresses the challenge of producing stable and faithful 3D edits without model finetuning for 3D content creation, representing an incremental improvement over existing methods.
The paper tackled the problem of training-free 3D editing, which often suffers from inconsistent latent anchors leading to weak or unstable edits, and introduced AnchorFlow to enforce latent anchor consistency, resulting in semantically aligned and structurally robust edits as shown on the Eval3DEdit benchmark.
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.