ComfyUI-FastSDCPU

ComfyUI-FastSDCPU
★ 11

模型推理远程调度ComfyUI扩展FastSDCPU集成
在 ComfyUI 中与 FastSDCPU Web 服务器通信的一组节点，方便将 FastSDCPU 的本地或远程推理能力集成到 ComfyUI 工作流中。
💡 在 ComfyUI 流程中调用 FastSDCPU 进行图像生成推理。
🍴 1 Forks💻 Python🔄 2024-09-16
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/e58c8376a81b
📦 requirements.txt
requests
pillow
numpy
torch
📄 README

ComfyUI-FastSDCPU

A set of nodes for interfacing with the FastSDCPU webserver

Setup

  • Install the extension with dependencies
  • Install FastSDCPU (clone the repo, run install.bat/install.sh)
  • (Recommended) Set the environment variable DEVICE=GPU:

    • Bash: export DEVICE=GPU

    Powershell: $env:DEVICE = "GPU"

    CMD: set DEVICE=GPU

  • Run start-webserver.bat/start-webserver.sh

    • Converting your own models

    Fastsdcpu supports regular diffusers models as well as openvino models. Openvino models are much faster than standard models on intel cpus and igpus but they also require significantly more memory. To convert them you can use the lcm-openvino-converter repo. This repo contains commands to both apply an lcm lora and convert the model to openvino. The lcm conversion isn’t required but because fastsdcpu is optimised around LCM models regular models will use inefficient sampling parameters which will require more steps than normal.

    Custom models should be placed into comfyui’s diffusers folder.

    Selecting local models only works when the webserver is run on the same system as the comfyui instance. If the server is not local you will have to move the folder to the server and specify the folder path that’s on the server.

    Previews