Skip to content

README.md

Latest commit

 

History

History
66 lines (44 loc) · 1.74 KB

README.md

File metadata and controls

66 lines (44 loc) · 1.74 KB

ComfyUI-Direct3D-S2

Direct3D-S2 ComfyUI Wrapper (Python v3.10, v3.11 or v3.12) Supported Models: 1.0 and 1.1

results

Can be used to refine existing mesh from Hy3D 2.0/2.1 etc or construct dense mesh from image.

VRAM Requirements

  • 8GB VRAM: Confirmed working (CPU offloaded, very slow)
  • 12GB VRAM: Minimum Recommended (Slow CPU offloading in complex cases)
  • 16–24GB VRAM: Ideal for best performance

Installation

Install requirements

cd ComfyUI\custom_nodes\ComfyUI-Direct3D-S2
..\..\..\..\python_embeded\python.exe -m pip install -r requirements.txt

Install voxelize

cd ComfyUI\custom_nodes\ComfyUI-Direct3D-S2\voxelize
..\..\..\..\python_embeded\python.exe setup.py install

Install torchsparse

Linux: pip install torchsparse

Windows: You will find wheels in the folder wheels

Install Flash Attention

Linux: pip install flash_attn

Windows: You can find precompiled wheels here https://huggingface.co/lldacing/flash-attention-windows-wheel/tree/main

Download the models

You will find the models here: https://huggingface.co/wushuang98/Direct3D-S2/tree/main

Clone HuggingFace Repository

cd ComfyUI/models
mkdir -p wushuang98
cd wushuang98
git clone https://huggingface.co/wushuang98/Direct3D-S2

Final structure should look like:

ComfyUI/models/wushuang98/Direct3D-S2/direct3d-s2-v-1-0/
ComfyUI/models/wushuang98/Direct3D-S2/direct3d-s2-v-1-1/

Make sure to place the respective checkpoints and config files inside each version folder.

Known Bugs

Occasional OOM's