Research map / Image generation / Editing and style transfer
AnchorFlow: Training-Free 3D Editing via Latent Anchor-Aligned Flows
Zhenglin Zhou, Fan Ma, Chengzhuo Gui, Xiaobo Xia, Hehe Fan, Yi Yang, Tat-Seng Chua
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace timestep-wise random anchors with anchor-aligned update | Hunyuan3D 2.1 inversion-free editing · Eval3DEdit (overall) | CLIP-txt: 0.4705 -> 0.4866 | Improved |
| replace timestep-wise random anchors with anchor-aligned update | Hunyuan3D 2.1 inversion-free editing · Eval3DEdit (overall) | CLIP-img: 0.7106 -> 0.7173 | Improved |
| anchor-aligned update on object removal | inversion-free editing · Eval3DEdit, Object Removal | CLIP-img / CLIP-txt: CLIP-img 0.7293 -> 0.7167; CLIP-txt 0.4884 -> 0.4735 | Hurt |
| anchor-aligned update on object addition | inversion-free editing · Eval3DEdit, Object Addition | CLIP-txt: 0.4441 -> 0.4758 | Improved |
| use direct editing from the edited image instead of latent-space editing methods | Hunyuan3D 2.1 · Eval3DEdit (overall) | CLIP-img: 0.7173 -> 0.6152 | Hurt |
| average multiple noisy flow directions (n_avg) (ablation) | AnchorFlow and inversion-free editing · Eval3DEdit | editing quality | Improved |
| increase n_max and s_tar (35,5.0 to 41,7.5) (ablation) | AnchorFlow · Eval3DEdit | CLIP-txt vs CLIP-img | Mixed |
| larger n_max and s_tar for action change; smaller for removal and style (ablation) | AnchorFlow · Eval3DEdit | CLIP-img and CLIP-txt | Improved |
| use anchor-aligned update instead of inversion-free update (runtime) | AnchorFlow · Eval3DEdit time | time (s): 25.77 -> 26.71 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports AnchorFlow as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- TanGO: Training-Free 3D Editing via Tangent-Space Guidance and Optimization (ECCV 2026 · better on 7 of 7 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Training-Free 3D Editing via Feature-Divergence Localization and Trajectory Correction (NeurIPS 2026 · similarity 0.88)
- TanGO: Training-Free 3D Editing via Tangent-Space Guidance and Optimization (ECCV 2026 · similarity 0.87)
- Velocity-Space 3D Asset Editing (NeurIPS 2026 · similarity 0.87)
- SyncNoise: Geometrically Consistent Noise Prediction for Instruction-based 3D Editing (AAAI 2025 · similarity 0.86)
- ConsistDreamer: 3D-Consistent 2D Diffusion for High-Fidelity Scene Editing (CVPR 2024 · similarity 0.86)
- SHAP-EDITOR: Instruction-Guided Latent 3D Editing in Seconds (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
