Guides
TRELLIS 2 for Mac: what runs on Apple Silicon, what doesn't, and what to use instead
Can TRELLIS 2 run on a Mac? Not officially. Here are the community MLX and MPS ports for Apple Silicon, what they need, and the hosted alternatives.
5 min read

Quick answer
Can you run TRELLIS 2 on a Mac?
Not officially. Microsoft tests it only on Linux with a 24 GB NVIDIA GPU. Community ports for Apple Silicon do exist, including a native MLX port tested on an M4 Max with 128 GB, but they're small hobby projects. For reliable results on a Mac, use TRELLIS 2 through a browser: the Hugging Face demo, Krea or Vi3W.
At a glance
| Route | Who makes it | Runs on | Output | Status | ||
|---|---|---|---|---|---|---|
| TRELLIS.2 official | Microsoft | Linux, NVIDIA 24 GB+ only | PBR GLB | No Mac support | ||
| gtrg55/trellis2-mlx | Community (MIT) | Apple Silicon via MLX; tested on an M4 Max with 128 GB | Textured GLB, 1024³ generation | Small hobby project; experimental | ||
| lyonsno/trellis2mlx | Community | Macs with more than 16 GB | Not stated | Experimental | ||
| sanchez-kim/trellis-silicon | Community | Apple Silicon via MPS and Metal | Not stated | Updated Nov 17, 2026; experimental | ||
| dmvvilela/ComfyUI-TrellisMac | Community | ComfyUI on Mac | Not stated | Experimental | ||
| Hosted TRELLIS 2 (HF Space | Krea | Vi3W) | Microsoft demo; vendors | Any Mac with a browser | GLB and other formats | Works today; usage limits vary |
Why there's no official Mac version
TRELLIS.2 is Microsoft's 4B-parameter image-to-3D model, MIT-licensed, which writes a PBR GLB from one image. Its README says it's tested only on Linux and needs an NVIDIA GPU with at least 24 GB, verified on A100 and H100.
Macs don't have NVIDIA GPUs, and the reference code is written for CUDA. So out of the box, it won't run on a Mac at all. Not slowly: not at all. Anything that works on Apple Silicon is a port, rewritten by someone in the community to use Apple's own frameworks instead.
We haven't benchmarked any of these ports, so there are no speed or quality figures on this page. What follows is what each project's own page says.
The community ports
gtrg55/trellis2-mlx: native MLX
The most complete-sounding port. It's MIT-licensed, rewritten for MLX, Apple's machine-learning framework, and its README says it does 1024³ generation and outputs a textured GLB. It was tested on an M4 Max with 128 GB of unified memory. That's a high-end machine; the README doesn't promise the same on a smaller Mac. It has around 70 stars on GitHub, which tells you it's a small project.
lyonsno/trellis2mlx
Another MLX port, aimed at Macs with more than 16 GB of memory. Treat that as the floor it targets, not a guarantee.
sanchez-kim/trellis-silicon: MPS and Metal
This one uses PyTorch's MPS backend and Metal rather than MLX. It was updated on November 17, 2026, so it's active.
dmvvilela/ComfyUI-TrellisMac
A ComfyUI wrapper for Mac users. If you already run ComfyUI on your Mac, it's the shortest path to trying TRELLIS in a node graph. Note that ComfyUI's own native TRELLIS.2 nodes (v0.34+) are designed around NVIDIA hardware, and their announcement doesn't mention the Mac.
All four are small hobby projects. Expect install steps that assume you're comfortable in Terminal, and expect breakage when macOS, Python or the upstream model changes. Read each README before you install, and keep a note of the version that worked.

Before you install a port
A short checklist before you start:
- Check your memory. About This Mac shows it. Compare it with what the port's README was tested on.
- Check the last commit. A port that hasn't moved since the upstream model changed may not load current weights.
- Read the open issues. Search for your chip and memory size. Someone may have already hit your problem.
- Use a separate Python environment. These projects pin their own versions, and mixing them with other AI tools breaks things.
- Read the licence. The ports listed here are MIT or unstated; the TRELLIS.2 weights are MIT.
What your Mac's memory means here
Apple Silicon shares one pool of unified memory between the CPU and GPU. That's why the ports talk about total memory (16 GB, 128 GB) rather than VRAM. The official model wants 24 GB of GPU memory on NVIDIA, and the best-documented Mac port was tested with 128 GB. If your Mac has 16 GB or less, a port may not fit at all; check the project's issues page for reports from machines like yours.
The reliable routes from a Mac
Use it in the browser
- Hugging Face Space: Microsoft links a TRELLIS.2 demo from the repo. It's shared, so limits apply.
- Krea: hosts TRELLIS 2 in its 3D feature, with 100 free compute units a day. Free-tier commercial terms are unclear, so check.
- Vi3W: runs TRELLIS and TRELLIS 2 in the browser alongside Hunyuan 3D, Meshy, Tripo, Rodin and others, 25 image-to-3D models in all. TRELLIS is the free default; other models cost XP, shown before each run.
Use an open model that lists macOS
Hunyuan3D 2.0's README lists macOS, Windows and Linux, with about 6 GB for shape and 16 GB with texture. Its licence isn't valid in the EU, the UK or South Korea. Walkthrough: Hunyuan 3D on Mac. Stable Fast 3D lists Mac support through MPS as experimental.
Rent a GPU
If you need TRELLIS.2 at full resolution with the official code, a Linux cloud machine with a 24 GB+ NVIDIA GPU runs it exactly as Microsoft tested it. Setup: run TRELLIS locally, which works the same on a rented box.
Where your Mac does shine: after the mesh
Whichever route makes the model, the Mac is a good place to finish it. Blender renders on Apple GPUs through Metal in Cycles, and Blender 5.3 (November 10, 2026) imports Gaussian splats natively. Import the GLB, check the wireframe and the texture maps, and remember that TRELLIS.2 exports in OPAQUE mode by default, so wire any alpha to the material yourself.
Which route to pick
- You want models today: a browser route.
- You have a high-memory Apple Silicon Mac and enjoy tinkering: gtrg55/trellis2-mlx.
- You want offline and open, and the licence fits your country: Hunyuan3D 2.0.
- You need the official code at full resolution: a rented Linux GPU.
For more Mac options beyond TRELLIS, see best AI 3D generator for Mac for 2027.
TRELLIS 2 without the port
Open Vi3W in your Mac's browser, upload an image, run it through TRELLIS 2 or another model, and download a GLB, FBX, OBJ, STL, PLY or glTF. Try Vi3W.
Frequently asked questions
trellis 2 for mac
There's no official Mac build. Your choices are a community port (gtrg55/trellis2-mlx, lyonsno/trellis2mlx, sanchez-kim/trellis-silicon or ComfyUI-TrellisMac), all experimental, or a hosted version you use in the browser.
trellis 2 on mac
If you have a high-memory Apple Silicon Mac and like tinkering, try the MLX port first, since it's native and outputs textured GLBs. If you just want models, use a hosted TRELLIS 2 and spend your Mac's power on Blender instead.
trellis 2 free demo
Microsoft's TRELLIS.2 repo links a Hugging Face Space demo, which runs in any browser, including on a Mac. It's shared, so check its current limits. Krea also hosts TRELLIS 2 with 100 free compute units a day.
Last reviewed

