Hunyuan3D-2
High-Resolution 3D Assets Generation with Large Scale Hunyuan3D Diffusion Models.
Tencent's open-source tool that converts a single photograph into a detailed 3D model with realistic textures, runnable locally on a GPU or via a hosted web studio without any setup.
Hunyuan3D-2 is a project from Tencent that turns ordinary images into three-dimensional objects. You give it a photograph or illustration, and it produces a 3D model complete with detailed geometry and realistic surface textures. The system is designed to work at high resolution, meaning the resulting models have fine detail rather than blocky, low-quality shapes.
Under the hood, the system works in two separate steps. First, a shape model called Hunyuan3D-DiT figures out the underlying geometry of the object in the image. Then a texture model called Hunyuan3D-Paint applies realistic color and surface detail to that geometry. Splitting the work this way means each part can be improved independently, and it also lets you apply new textures to 3D models you already have, not just ones generated from scratch.
Tencent has released the model weights and code publicly, so anyone with the right hardware can run it locally. There is also a hosted web studio at the official Tencent site where you can try it without setting anything up. The community has added support for ComfyUI, a visual workflow tool popular with image-generation enthusiasts, and a Windows installer for people who do not want to handle the setup manually. A Blender addon is available as well, letting 3D artists feed images into the system directly from inside Blender.
Performance comparisons in the README show Hunyuan3D 2.0 scoring better than several competing models, both open-source and commercial, across standard image-quality metrics and how closely the output matches the input image. Faster turbo variants of both the shape and texture models have since been released to reduce the time it takes to generate results.
The project continues to evolve. Newer versions have introduced a physically-based rendering texture model, a multi-view shape model that uses more than one camera angle as input, and a smaller 0.6B-parameter variant for machines with less memory. There is also a 3D world generation model called HunyuanWorld that the team released as a separate project more recently.
Where it fits
- Turn a product photo into a 3D model for use in a game, e-commerce listing, or design mockup.
- Apply a new texture to an existing 3D mesh without regenerating the geometry from scratch.
- Run image-to-3D generation inside Blender using the provided addon for direct workflow integration.
- Try 3D model generation in a browser via the hosted Tencent web studio without installing anything.