In the field of 3D modeling, generating realistic textures has always been a challenging task. Recently, researchers from the University of Alberta, the University of Toronto, and Huawei's Noah's Ark Lab have jointly introduced a new method — TexGen. This technology can generate corresponding 3D textures based on user text descriptions, and its effectiveness significantly surpasses existing similar technologies.

image.png

Product Entry: https://top.aibase.com/tool/texgen

The core of TexGen is a multi-view sampling and resampling framework, which operates with a pre-trained text-to-image diffusion model. Traditional methods often result in noticeable seams or excessive smoothing when generating textures.

TexGen maintains a texture map parameterized in RGB space and updates it after each sampling step of the diffusion model, effectively reducing inter-view differences. This view-consistent sampling method makes the textures between each view more coherent and natural.

image.png

TexGen also introduces an attention-guided multi-view sampling strategy, which allows appearance information of textures to be broadcast across different views, ensuring consistency across all views.

Additionally, to preserve texture details, the research team developed a noise resampling technique, which helps estimate noise and generate suitable inputs for subsequent denoising steps. This way, TexGen not only achieves breakthroughs in texture generation but also excels in texture editing, allowing adjustments while retaining the original identity.

After extensive qualitative and quantitative evaluations, the research team found that TexGen performs exceptionally well in generating high-quality, view-consistent, and richly detailed textures for diverse 3D objects, surpassing the current state-of-the-art methods.

Below is a comparison of TexGen with TEXTure, Text2Tex, Fantasia3D, and ProlificDreamer

image.png

image.png

Here are some demonstration effects:

image.png

image.png

image.png

Key Points:

🌟 TexGen can generate high-quality 3D textures based on text descriptions, addressing issues of seams and excessive smoothing in traditional methods.

👀 Utilizes a multi-view sampling and resampling framework to ensure texture consistency across views.

🖌️ New noise resampling technology makes TexGen excel in texture editing, preserving the original style.