Meta Unveils AssetGen 2.0: Next-Gen 3D Foundation Model 90

小编 · 2025-05-12

AssetGen 2.0 from Meta is a new 3D model generation tool that rapidly creates refined, realistic 3D models and textures from text or image prompts. Built on single-stage 3D diffusion technology, it streamlines the generation process while keeping geometry and textures more consistent. Meta's vision is to make creating 3D content as easy as creating 2D content, empowering users to build virtual worlds. Looking ahead, Meta is also working on a new feature that will generate an entire 3D scene in a single pass.

Meta Unveils AssetGen 2.0: Next-Gen 3D Foundation Model

What is AssetGen 2.0?

AssetGen 2.0, introduced by Meta, is a new 3D model generation tool. It can quickly produce refined and realistic 3D models and textures from text or image prompts. AssetGen 2.0 uses single-stage 3D diffusion technology, making the generation process simpler while keeping the model's shape and texture more consistent. Meta aims to make creating 3D content as easy as creating 2D content with this tool, helping everyone build virtual worlds. In addition, Meta is also working on a new feature that will be able to generate entire 3D scenes at once, making 3D content creation more accessible.

Meta launches next-generation 3D foundation model: AssetGen 2.0.webp

AssetGen 2.0 Features

AssetGen 2.0 is a generative model specifically designed to convert text and image prompts into high-quality 3D assets. It is composed of two sub-models developed internally by Meta: one for generating 3D meshes, and another for generating textures.

Technical Architecture

Single-stage 3D diffusion model: Unlike AssetGen 1.0, AssetGen 2.0 uses a single-stage 3D diffusion model to generate 3D meshes directly from text prompts, rather than first generating 2D views and then processing them. This architecture enables it to generate 3D assets with geometric consistency and fine details.
Improved texture generation: AssetGen 2.0 includes an upgraded TextureGen model for generating high-quality textures. The new model brings improvements in view consistency, texture inpainting, and texture resolution.

Use Cases

3D world creation: Meta is currently using AssetGen 2.0 internally to create 3D virtual worlds, and plans to roll it out to Horizon creators and the Meta Horizon Desktop editor later this year.
Full 3D scene generation: Meta also plans to launch a feature for generating full 3D scenes based on AssetGen 2.0 in the coming months, enabling creators to quickly generate complex virtual environments from simple text or image prompts.

What improvements does AssetGen 2.0 have over AssetGen 1.0?

Building on the foundation of AssetGen 1.0, AssetGen 2.0 represents a major breakthrough in 3D generative AI research. It adopts a single-stage 3D diffusion model, significantly improving the detail expression and realism of 3D meshes. The accompanying TextureGen model ensures that the assets are not only visually striking but also feature high-quality textures that meet production environment needs. The model maintains consistency in geometric structure while delivering extremely fine details. Unlike AssetGen 1.0, it uses a 3D diffusion mechanism for geometric modeling and is trained on a large-scale 3D asset corpus. For texture generation, it introduces new techniques such as enhanced view consistency, texture inpainting, and higher-resolution texture generation.

In addition, the 3D assets generated by AssetGen 2.0 show more noticeable improvements in detail and fidelity compared to AssetGen 1.0, and it is capable of generating assets with fine geometric details.

Details:https://developers.meta.com/horizon/blog/AssetGen2