Tencent-Hunyuan/Hunyuan3D-2.1

Features

• Image-to-Mesh Generation - Provides a complete pipeline for converting single 2D images into three-dimensional polygonal mesh representations.
• Flow-Matching Frameworks - Utilizes a flow-matching pipeline to transform Gaussian noise into initial 3D asset shapes.
• Texture Synthesis - Produces photorealistic PBR maps by iteratively denoising image latent spaces conditioned on mesh geometry.
• 3D Asset Generators - Functions as a generative framework that synthesizes 3D meshes and textures from image inputs.
• Image-Conditioned 3D Generation - Implements a system where 2D image features guide the flow-matching process to synthesize matching 3D geometries.
• Generative PBR Textures - Synthesizes photorealistic material maps including metallic, roughness, and albedo properties for realistic light simulation.
• Single-Image 3D Reconstructions - Implements a flow-matching system to reconstruct a complete 3D digital model from a single photograph.
• Latent Space Generative Models - Compresses complex 3D geometric data into lower-dimensional vectors to accelerate generation and sampling.
• 3D Asset Pipelines - Automates the end-to-end conversion of flat images into fully textured 3D models for digital environments.
• Iterative Geometry Refinement - Transforms initial coarse shapes into high-fidelity geometries through a sequence of iterative optimization and smoothing passes.

Star history