ComfyUI-Trellis2Apple

ComfyUI-Trellis2Apple

Experimental ComfyUI nodes for pedronaugusto/trellis2-apple, using its MLX backend on Apple Silicon.

This is a separate project from ComfyUI-TrellisMac. Keep both installed if you want to compare outputs.

Status

Experimental. Direct MLX generation/export has been smoke-tested with the same Python environment that launches ComfyUI. The ComfyUI workflow still needs GUI validation.

Install

1. Clone Pedro's backend repo:

git clone https://github.com/pedronaugusto/trellis2-apple /path/to/trellis2-apple

2. Put this repo in ComfyUI's custom_nodes/:

ln -s /path/to/ComfyUI-Trellis2Apple /path/to/ComfyUI/custom_nodes/ComfyUI-Trellis2Apple

3. Install with the same Python that launches ComfyUI:

/path/to/ComfyUI/.venv/bin/python custom_nodes/ComfyUI-Trellis2Apple/install.py \
  --trellis2-apple /path/to/trellis2-apple \
  --download-weights

4. Verify:

/path/to/ComfyUI/.venv/bin/python custom_nodes/ComfyUI-Trellis2Apple/doctor.py

Notes

• This backend uses mlx and Pedro's MLX adapters.
• It currently expects local weights at trellis2-apple/weights/TRELLIS.2-4B.
• Pedro's macOS requirements pin transformers<5; that may affect other custom nodes in the same ComfyUI Python environment.
• The example workflow starts with the direct-tested safer settings: 512 - fast, 300000 decimation target, 1024 texture size.
• Loader fields can stay as auto; the node reads config.json and uses local weights from the trellis2-apple checkout.
• The standard Generate node only exposes seed and pipeline type. Use the Advanced Generate node if you want sampler guidance/steps.
• Use GLB for ComfyUI Preview3D.

Nodes

LoadImage -> Load Trellis2 Apple MLX Pipeline -> Trellis2 Apple MLX Image to 3D -> Export Trellis2 Apple Mesh -> Preview3D

Credits

• TRELLIS.2 by Microsoft Research
• trellis2-apple by Pedro Naugusto