We propose OmniStyle2 that reframes artistic style transfer through destylization, a process that removes stylistic elements from artworks to recover clean, style-free content. By constructing the large-scale DST-100K dataset with this approach, we enable a simple feed-forward model that consistently surpasses state-of-the-art methods.

We propose TexVerse, a large-scale 3D dataset featuring high-resolution textures. TexVerse collects over 858K unique high-resolution 3D models sourced from Sketchfab, including more than 158K models with physically based rendering (PBR) materials.

We propose DecoupledGaussian that separates static objects from their contacted surfaces in in-the-wild videos, enabling realistic object-scene interactive simulations.

We propose Diff3DS, a novel differentiable rendering framework for generating view-consistent 3D sketch. Diff3DS enables end-to-end optimization of 3D sketches via gradients in the 2D image domain, supporting novel tasks like text-to-3D sketch and image-to-3D sketch.

We propose 3D-SceneDreamer that provides a unified solution for text-driven 3D consistent indoor and outdoor scene generation.