Tag
This paper introduces a feed-forward 3D editing framework that uses generative prior distillation from foundation models to avoid paired training data, improving instruction fidelity and cross-view consistency.