Method
The overview of TexGenerator pipeline.
Illustrations of the U-Net architectures in (a) Latent Diffusion Model
TexGenerator, Our mesh texture stylization method enables high quality texture style transfer from a input style image while preserve the original content of mesh's texture using the pretrained diffusion model.
In this paper, we introduce TextureGenerator, an innovative framework designed to produce high-quality textures for 3D Meshes from the given reference images. Our approach introduces a novel framework for few-shot style personalization of diffusion models by employing a combination of Stable diffusion XL, Low Ranked Adapters (LoRA),ControlNets and IP-Adapter,These components are cascaded during training to produce a personalized diffusion model tailored to the reference images, for cases where the Pre-trained ControlNet and LORA parameters affect each other, and the pre-trained IP-Adapter also fails to capture image details from the input reference image . Additionally,We integrate a multi-view diffusion model to generate precise 3D textures that ensure both style and semantic consistency. Furthermore, we also propose an automatic multi-view texture compositing technique, which ensures the compositing of high-quality texture.Compared to the current state-of-the-art methods, TextureGenerator significantly enhances the similarity and quality of textures generated from reference images, providing strong support for the creation of realistic, high-quality 3D assets.
The overview of TexGenerator pipeline.
Illustrations of the U-Net architectures in (a) Latent Diffusion Model
Please consider citing our work if you find it useful.
@article{TexGenerator,
title = {{TexGenerator: Advancing High-Quality Texture Creation with Image-Guided Diffusion Models}},
author = {},
journal = {arxiv preprint arXiv:},
year = {2024},
}