Feed-forward 3D editing without paired 3D supervision, guided by visual, semantic, and geometric priors.
Haoyuehx © 2026