ComfyUI-3D-Pack

repository·main·Indexed 25 days ago

https://github.com/mrforexample/comfyui-3d-pack

An extensive node suite for ComfyUI version 0.1.6 that enables 3D asset generation. It supports processing 3D inputs such as Mesh and UV Textures using 3DGS and NeRF algorithms, and integrates models including InstantMesh, CRM, TripoSR, CharacterGen, TRELLIS, and Hunyuan3D-1.0.

Tokens: 12.1K
Snippets: 30
Records: 75
Agent score: 87%

What's inside ComfyUI-3D-Pack

Overview of TRELLIS Normal Conditioned Model
main
The trellis-normal-v0-1 model is an improved version of the TRELLIS image-to-3d pipeline that utilizes normal conditioning. It is designed to enhance the quality of 3D generation by leveraging surface normals.
https://github.com/mrforexample/comfyui-3d-pack/blob/main/Checkpoints/Diffusers/Stable3DGen/trellis/trellis-normal-v0-1/README.md
Overview of CharacterGen model parameters
main
This directory contains the model parameters for CharacterGen, an image-to-3d pipeline designed for efficient 3D character generation from single images using multi-view pose canonicalization.
https://github.com/mrforexample/comfyui-3d-pack/blob/main/Checkpoints/CharacterGen/README.md
Overview of PartCrafter for Structured 3D Mesh Generation
main
PartCrafter is a structured 3D generative model designed to jointly generate multiple parts and objects from a single RGB image in a single shot. It utilizes compositional latent diffusion transformers. This specific repository contains the pretrained model weights for PartCrafter at the object level.
https://github.com/mrforexample/comfyui-3d-pack/blob/main/Checkpoints/Diffusers/wgsxm/PartCrafter/README.md
Overview of PartCrafter Scene-Level 3D Generation
main
PartCrafter is a structured 3D generative model designed for scene-level generation. It can jointly generate multiple parts and objects from a single RGB image in a single shot using Compositional Latent Diffusion Transformers. This repository provides the pretrained model weights for the scene level implementation.
https://github.com/mrforexample/comfyui-3d-pack/blob/main/Checkpoints/Diffusers/wgsxm/PartCrafter-Scene/README.md
Overview of TRELLIS Image Large
main
TRELLIS Image Large is an image-conditioned version of the TRELLIS large 3D generative model. It is designed for image-to-3d pipelines, allowing for the generation of 3D assets from 2D images based on the research presented in Structured 3D Latents for Scalable and Versatile 3D Generation.
https://github.com/mrforexample/comfyui-3d-pack/blob/main/Checkpoints/Diffusers/jetx/TRELLIS-image-large/README.md
Overview of TripoSG Model
main
TripoSG is an image-to-3D generation foundation model that uses large-scale rectified flow transformers to generate high-fidelity 3D shapes from a single input image. It is intended for converting single images to 3D meshes, gaming/VFX asset creation, and prototyping.
https://github.com/mrforexample/comfyui-3d-pack/blob/main/Checkpoints/Diffusers/VAST-AI/TripoSG/README.md
Use TripoSG-scribble for 3D shape prototyping
main
TripoSG-scribble is a variant of TripoSG designed for fast 3D shape prototyping. It converts a scribble image and a text prompt into a 3D mesh.
Critical Inference Setting: Because TripoSG-scribble is a CFG-distilled model, you must use CFG=0 during inference to ensure correct results.
https://github.com/mrforexample/comfyui-3d-pack/blob/main/Checkpoints/Diffusers/VAST-AI/TripoSG-scribble/README.md
Project structure overview for developers
main
For developers looking to extend the pack, the following directory structure is used:
- nodes.py: Interface code for all ComfyUI nodes. Add new nodes here.
- Gen_3D_Modules/: Code for generative models (e.g., multi-view diffusion, 3D reconstruction).
- MVs_Algorithms/: Multi-view stereo algorithms (e.g., Gaussian Splatting, NeRF, FlexiCubes).
- web/: Browser UI files (HTML, JS, CSS) for 3D mesh/Gaussian previews.
- webserver/: Web server logic for handling client requests and sending 3D data.
- Configs/: Configuration files for various modules (e.g., Configs/Unique3D_configs).
- Checkpoints/: Pre-trained models and architecture configs. Files here can be automatically downloaded by the Load_Diffusers Pipeline node.
- install.py: Main installation script that handles pre-builds and automatic builds via ComfyUI-Manager.
- _Pre_Builds/: Files and code used to build required dependencies.
https://github.com/mrforexample/comfyui-3d-pack/blob/main/README.md
Install and run ComfyUI-3D-Pack with Docker (Linux)
main
To run Comfy3D in a Docker container on Linux, you must have GPU support configured via nvidia-container-runtime.
1. Install Runtime: sudo apt-get install nvidia-container-runtime.
2. Configure Docker: Edit or create /etc/docker/daemon.json to include the NVIDIA runtime as the default:
  { "runtimes": { "nvidia": { "path": "/usr/bin/nvidia-container-runtime", "runtimeArgs": [] } }, "default-runtime": "nvidia" }
3. Restart Docker: sudo systemctl restart docker.
4. Build and Run: Use docker build and docker run with the --gpus all flag.
```
# Configure daemon.json first, then:
sudo systemctl restart docker
docker build -t comfy3d . && docker run --rm -it -p 8188:8188 --gpus all comfy3d
```
https://github.com/mrforexample/comfyui-3d-pack/blob/main/_Pre_Builds/README.md

Run the Comfy3D Docker container

main

You can run the container using docker run with specific volume mappings for models and outputs, or simply use docker compose up.

# Using docker run
docker run -d \
  --name comfy3d \
  --platform linux/amd64 \
  --runtime=nvidia \
  -p 8188:8188 \
  -e NVIDIA_VISIBLE_DEVICES=all \
  -e NVIDIA_DRIVER_CAPABILITIES=all \
  -e COMFYUI_PATH=/app \
  -v $(pwd)/output:/app/output \
  -v $(pwd)/models/diffusion_pytorch_model.safetensors:/app/models/controlnet/control_unique3d_sd15_tile.safetensors \
  -v $(pwd)/models/ip-adapter_sd15.safetensors:/app/models/ipadapter/ip-adapter_sd15.safetensors \
  -v $(pwd)/models/RealESRGAN_x4plus.pth:/app/models/upscale_models/RealESRGAN_x4plus.pth \
  -v $(pwd)/models/v1-5-pruned-emaonly.ckpt:/app/models/checkpoints/v1-5-pruned-emaonly.ckpt \
  -v $(pwd)/models/model.safetensors:/app/models/clip_vision/OpenCLIP-ViT-H-14.safetensors \
  comfy3d

# Using Docker Compose
docker compose up

Set up Blender requirements for rendering
main
The rendering scripts require Blender 4.1 and specific Python packages (opencv, OpenEXR, and Imath) installed within Blender's internal Python environment.
On macOS, you can install these dependencies using the following command (adjusting the path to your Blender installation if necessary):
```
/Applications/Blender.app/Contents/Resources/4.1/python/bin/python3.11 -m pip install OpenEXR Imath opencv-python
```
https://github.com/mrforexample/comfyui-3d-pack/blob/main/Gen_3D_Modules/Hunyuan3D_2_1/hy3dshape/tools/README.md
Manual Installation of ComfyUI-3D-Pack
main
To install manually, clone the repository into your ComfyUI custom nodes directory and run the provided installation scripts using your ComfyUI embedded Python executable. This process attempts to download pre-built binaries for your environment or triggers an automatic build if pre-builds are unavailable.
```
# Fetch newest version of Comfy3D
cd Your ComfyUI Root Directory\ComfyUI\custom_nodes\
git clone https://github.com/MrForExample/ComfyUI-3D-Pack.git
cd ComfyUI-3D-Pack

# Install all dependencies
Your ComfyUI Root Directory\python_embeded\python.exe -s -m pip install -r requirements.txt
Your ComfyUI Root Directory\python_embeded\python.exe install.py
```
https://github.com/mrforexample/comfyui-3d-pack/blob/main/README.md