Hunyuan3D-2 on Apple Silicon

The honest state of Mac support today, and what is coming to 3D Mate.

Hunyuan3D-2 is Tencent's high-detail image-to-3D model. It is powerful but heavy, and its reference code is built for NVIDIA GPUs, so there is no simple way to run the full pipeline on Apple Silicon today. Here is the honest state of Mac support, the memory it wants, and how 3D Mate is bringing it on-device.

The manual route

1

What Hunyuan3D-2 needs

The pipeline has a shape stage and a separate texture stage. Both are large and CUDA-oriented, and the texture pass is especially memory hungry.

2

Mac support today

Running the official code on a Mac means porting its CUDA components to Apple's MLX or PyTorch MPS, or renting a cloud GPU. There is no first-party Apple Silicon build.

3

The memory reality

Even ported, the geometry pass wants around 24 GB of unified memory and the texture pass around 48 GB, so it favours a high-memory Mac.

Hunyuan3D-2 is the most demanding engine here, and today it does not run natively on a Mac without a substantial port.

The one-click way

Hunyuan3D-2 is coming to 3D Mate as a Mac-native, offline engine. Until then, TRELLIS.2 4B (balanced) and TripoSG (sharpest geometry) run fully offline on Apple Silicon today and cover most needs.

Questions

Can I run Hunyuan3D-2 on a Mac right now?
Not from the official CUDA code without a major port. A Mac-native version is coming to 3D Mate; meanwhile TRELLIS.2 4B and TripoSG run offline on Apple Silicon today.
How much memory will Hunyuan3D-2 need?
Around 24 GB of unified memory for geometry and about 48 GB for the texture pass, so a high-memory Mac is best.
Is it available in 3D Mate yet?
It is marked coming soon. The shipping engines cover balanced quality (TRELLIS.2 4B) and sharp geometry (TripoSG) today.