Meta Unveils Revolutionary AI Model Meta3DGen for 3D Content Generation

TapTechNews July 3rd news, Meta company yesterday (July 2nd) released a research paper introducing a brand-new AI model called Meta3DGen (3DGen), which can generate high-quality 3D content within 1 minute based on the prompt words input by users.

Meta company stated that the 3D content generated by Meta3DGen (3DGen) has high-resolution textures and material mappings, and also supports physically based rendering (PBR), and can regenerate new textures for the previously generated 3D content.

TapTechNews quoted the main content of the paper. Meta3DGen includes two main components: text-to-3D generation and text-to-texture generation, and its working principle is as follows:

First stage

Based on the text prompt provided by users, the first stage uses Meta's 3D asset generation model (AssetGen) to create the initial 3D asset. This step generates a 3D mesh with textures and PBR material mappings. The reasoning time is about 30 seconds.

Second stage

Case 1: Generate more refined 3D textures

Based on the 3D asset generated in stage 1 and the initial text prompt, stage 2 will generate higher-quality textures and PBR mappings for this asset. It uses Meta's text-to-texture generator Meta3DTextureGen. The reasoning time is about 20 seconds.

Case 2: Regenerate 3D textures

Given an untextured 3D mesh and a prompt describing the desired appearance. Stage 2 can also generate textures for this 3D asset from scratch (the mesh can be previously generated or created by an artist). The reasoning time is about 20 seconds.

TapTechNews attached the complete report link, and interested users can read it in depth.