TanGO is a training-free framework for 3D editing on VecSet-based flow models. Instead of applying a single global update, it performs adaptive per-token steering in the tangent space of the generative dynamics — amplifying guidance for tokens in regions to be edited while attenuating it for tokens to be preserved — enabling mask-free, localized, identity-preserving edits. (Click the teaser below to see TanGO in action 👇)

Interactive 3D — drag any model to rotate. TanGO (Ours) preserves the source identity while applying a clean, localized edit. (More examples coming soon.)

This project webpage is built upon the TRELLIS and AnchorFlow project pages. We thank their authors for open-sourcing the webpage design and assets.

Citation will be available soon.