Skip to content

Latest commit

 

History

History
130 lines (91 loc) · 5.76 KB

README.md

File metadata and controls

130 lines (91 loc) · 5.76 KB

← Go back

ComfyUI Universal

A dockerized environment for running image generation AI's easily.⚡

Description • Requirements • Pre-installation Steps • Installation
Execution • Configuration • Storage • Known Issues • Credits

Description

This project is made for running ComfyUI easily on Docker, providing an easy configuration that can be adapted for image generation AI's. The default configuration and some parts of this README are based on the Flux.1-Schnell FP8 model.

Requirements

Required Recommended
GPU NVIDIA (with CUDA) NVIDIA (with CUDA)
GPU VRAM 4 GB 8 GB
CPU 4 cores/threads 8 cores/threads
DRAM 16 GB 32 GB
Disk space 30 GB 50 GB
CUDA Toolkit Required (not required for Docker) Required (not required for Docker)
NVIDIA Container Toolkit Required on Linux Required on Linux

Pre-installation Steps

Linux

  1. Install NVIDIA Container Toolkit:

  2. Verify NVIDIA driver version: Ensure that you are running the newest compatible version of the NVIDIA driver.

Windows

  1. Verify NVIDIA driver version: Ensure that you are running the newest compatible version of the NVIDIA driver.

Installation

  1. Clone the main repository:

    git clone https://github.com/TD99/ai-sandbox.git
cd ai-sandbox/comfyui-universal

  2. Install dependencies:

    docker-compose build

Execution

  1. Run ComfyUI:

    sudo docker-compose up [-d]

  2. Access ComfyUI: Open a web browser and navigate to http://localhost:8188/

  3. Load models in correct workflow: Click on "Load" in ComfyUI, select owl.json.

Configuration

models.json - Defines all model files that need to be loaded.

Property Description
name Filename
displayName Display name
description Short description
version Version
license License
origin URL from where to download
location Location in ComfyUI (use @ for root folder /app/ComfyUI)

.env - Defines settings.

Variable Description
PROJECT_ROOT Defines ComfyUI root folder
OVERRIDE_MODELS Always download models anew
MODEL_CONFIG_PATH Path to models.json
BEFORE_DEPLOY_CMD Command to run before deployment
AFTER_DEPLOY_CMD Command to run after deployment
TIMEOUT_RETRIES Number of retries for downloading model files

Storage

.models folder - All models are downloaded to the .models folder.

Known Issues

  • Windows Validation Issue: Validation sometimes fails in Windows.
    Solution: Change line endings from CRLF to LF in check_compatibility.sh, or set BEFORE_DEPLOY_CMD to empty in .env.

  • Model Download Issue: After downloading a model, no action occurs.
    Solution: Force quit the container using CTRL+C twice and start again.

Screenshots

ComfyUI in action

ComfyUI

Credits

Original Project: ComfyUI is a product of Comfy Org.

Dockerization and Optimization: This project was created by TD99 with contributions from other developers in the open-source community.

Codebase and Dependencies:
NVIDIA Container Toolkit: A product of NVIDIA Corporation.
Docker: A containerization platform developed by Docker Inc.