ComfyUI-AIT

ComfyUI-AIT
★ 51

加速推理AITemplateCUDA/C++稳定后端
基于Facebook Meta AITemplate的ComfyUI实现，利用cpp/cuda加速推理，提供更稳定的AI推理后端。
💡 用于在ComfyUI中通过AITemplate显著加速模型推理。
🍴 4 Forks💻 Python🔄 2024-06-22
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/86a6deb1b5f6
📄 README

ComfyUI-AIT

A ComfyUI implementation of Meta’s AITemplate repo for faster inference using cpp/cuda. The old AIT repo is still available for reference. You can find it here. The node in the UI is located under loaders->AIT. This new repo is behiond but is a much more stable foundation to keep AIT online. *Please be patient as the repo will eventually include the same features as before.*

Installation

This repo can be downloaded using ComfyUI manager.

To install the repo manually, simply git bash inside the custom_nodes directory and use the command git clone https://github.com/FizzleDorf/ComfyUI-AIT.git

Linux

  • open a terminal pathed to the current folder and use git clone --recursive https://github.com/facebookincubator/AITemplate.
  • path to cd python.
  • run python setup.py bdist_wheel.
  • If you are using a virtual environment, make sure it’s activated before installing the wheel.
  • pip install dist/*.whl --force-reinstall
  • copy and paste the 3rdparty directory to the Lib directory of your venv or python install.

    • Windows

    *This is for compilation only, you can do the Linux install for inference only*

  • Extract the ait_windows.zip file in \ComfyUI-AIT\compile directory.
  • open a cmd pathed to the current folder and use git clone --recursive ait_windows.bundle -b fixes.
  • path to cd ait_windows/python.
  • run python setup.py bdist_wheel.
  • If you cloned your ComfyUI install and you are using a virtual environment, make sure it’s activated before installing the wheel.
  • pip install dist/*.whl --force-reinstall for cloned comfy installs or .\python_embed\python.exe -s -m pip install for packaged comfy.
  • copy and paste the 3rdparty directory to the Lib directory of your venv or python install.

    • Modules

    You can find already compiled modules here otherwise you can compile your own modules (requires sm80 or higher NVIDIA GPU).

    *a bundle for Windows compilation is included with this repo. See Windows installation instructions.*

    Place your downloaded or compiles Modules in the ComfyUI/models/ait directory.

    While in ComfyUI, simply hook up the AIT_Unet_Loader to your workflow and select a module that is equal to or larger than the resolution you are generating at.