• Qinchan Li, New York University


• Finley Torrens,


• Kenneth Chen, New York University


• Qi Sun, New York University

Our BlendFusion generative 3D mesh modeling framework. (a) shows the overall and zoomed-in 3D scene for re-texturing, (b) shows the modeled texture after 20 minutes by a professional designer, and (c) shows the results using our framework with an inference time of 2 minutes with prompt `a child drawing style of a cabinet in a cartoony kitchen'.

Abstract

---

Modeling 3D assets is a universal in various applications, including animation and game development. However, a key challenge lies in the labor-intensive task of 3D texturing, where creators must repeatedly update textures to align with modified geometric shapes on the fly. This iterative workflow makes 3D texturing significantly more cumbersome and less efficient than 2D image painting. To address this, we introduce BlendFusion, an interactive framework that leverages generative diffusion models to streamline 3D texturing. Unlike existing systems that generate textures from scratch, BlendFusion integrates the procedural nature of texturing by incorporating multi-view projection to guide the generation process, enhancing stylistic alignment with the creator's intent. Experimental results demonstrate the robustness and consistency of BlendFusion across both objective and subjective evaluations.

Award

---

Citation

---

 @inproceedings{
 li2025blendfusion,
 author = {Li, Qinchan and Torrens, Finley and Chen, Kenneth and Sun, Qi},
 title = {BlendFusion: Procedural 3D Texturing Assistant with View-Consistent Generative Models},
 year = {2025},
 isbn = {9798400718335},
 publisher = {Association for Computing Machinery},
 address = {New York, NY, USA},
 url = {https://doi.org/10.1145/3722564.3728376},
 doi = {10.1145/3722564.3728376},
 booktitle = {Companion Proceedings of the ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games},
 articleno = {3},
 numpages = {3},
 series = {I3D Companion '25}
 }