TanGO: Training-Free 3D Editing via Tangent-Space Guidance and Optimization

2026-07-16 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

TanGO is a novel training-free framework designed for 3D editing, addressing semantic artifacts prevalent in recent flow-matching 3D generative models like VecSet. These models, despite using structured representations, suffer from issues such as collapsed preserved regions or incomplete transformations due to global context sharing among tokens during conventional editing. TanGO tackles this by enabling adaptive per-token steering within the tangent space of generative dynamics. The framework achieves selective control through a formulated one-step optimal control rule, which determines the strength of each token's control signal using a von Mises-Fisher inspired directional discrepancy derived from source and target velocity fields. Experimental results demonstrate that TanGO significantly reduces structural artifacts and achieves state-of-the-art performance, surpassing existing 3D editing baselines. Its code is publicly available, published on 2026-07-16.

Key takeaway

For 3D artists or ML engineers struggling with semantic artifacts in generative 3D model editing, TanGO offers a significant advancement. You should consider integrating this training-free framework to achieve cleaner, more complete transformations without global context issues. Its adaptive per-token steering capability means your edited 3D assets will exhibit substantially reduced structural artifacts, improving overall quality and workflow efficiency.

Key insights

TanGO enables training-free 3D editing by adaptively steering individual tokens in tangent space, reducing semantic artifacts.

Principles

• Adaptive per-token steering improves 3D editing.
• Tangent-space guidance resolves global context issues.
• Directional discrepancy informs control signal strength.

Method

TanGO formulates a one-step optimal control rule to enable selective per-token steering. It determines control signal strength using a von Mises-Fisher inspired directional discrepancy from velocity fields.

In practice

• Use TanGO for artifact-free 3D model transformations.
• Apply per-token steering to complex 3D scenes.
• Explore tangent-space guidance for generative models.

Topics

• 3D Editing
• Generative Models
• Flow-Matching
• Tangent Space
• Optimal Control
• VecSet

Code references

• siw00-lim/TanGO

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• Edit3DGS: Unified Framework for Dynamic Head Editing via 2D Instruction-Guided Diffusion and 3D Gaussian Splatting — Edit3DGS unifies 2D diffusion and 3D Gaussian splatting for dynamic, temporally consistent 3D head editing from video.
• Conflict-Aware Additive Guidance for Flow Models under Compositional Rewards — Gradient misalignment causes off-manifold drift when composing multiple constraints in guided sampling for flow models.
• ELSA3D: Elastic Semantic Anchoring for Unified 3D Understanding and Generation — ELSA3D unifies 3D understanding and generation by elastically anchoring language to geometry across abstraction scales.
• Compos3D: Interactive Part-Based Composition for Creative Control in Generative 3D Models — Compos3D enables greater creative control in generative 3D modeling through an interactive, part-based remixing workflow.
• DanceOPD: On-Policy Generative Field Distillation — DanceOPD unifies conflicting image generation capabilities by distilling expert velocity fields into a single flow-matching model.
• 1D-Bench: A Benchmark for Iterative UI Code Generation with Visual Feedback in Real-World — Iterative UI code generation with visual feedback improves performance, but advanced training methods need refinement.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.