Installing and Setting Up Comfy UI: A Comprehensive Guide
Comfy UI is a powerful tool that allows users to create custom workflows and explore the world of AI-generated art. In this guide, we will walk you through the steps to install and set up Comfy UI, as well as provide an overview of its features and capabilities.
Step 1: Download Comfy UI from GitHub
To get started, head over to the Comfy UI repository on GitHub and download the zip file. Once downloaded, extract the file using WinRAR or 7-Zip. This will give you access to the Comfy UI folder, which contains all the necessary files and information to get started.
Step 2: Understand the Readme File
Inside the extracted folder, you'll find a Readme file that provides important information to get you started with Comfy UI. If you have an Nvidia GPU, you'll want to use the .p file; otherwise, you can run Comfy UI on CPU, although this is not recommended as it will be slower. The Readme file also includes troubleshooting tips, such as ensuring you have a model checkpoint and updating Comfy UI by running the update command.
Step 3: Configure Comfy UI (Optional)
If you have an existing installation of Stable Fusion or Automatic111, you can configure Comfy UI to use your existing models. To do this, rename the config.example.json file to config.json and edit the file using Notepad. Set the base path to your Automatic111 installation, and save the file to enable Comfy UI to use your existing models.
Step 4: Download a Model (Optional)
If you don't have any models, you can download one from the CIVI website, such as the Deliberate model. Place the downloaded model in the models folder, and if you're unsure, put your regular save tensors or .ckpt files in the checkpoints folder.
Comfy UI Overview
Comfy UI is a powerful user interface that allows for total freedom and control in creating custom workflows. The default interface is composed of nodes, similar to other user interfaces like Automatic 1111. Nodes are connected to represent features, allowing users to create custom workflows. Multiple nodes can be added to create complex workflows.
Load Checkpoint/Model Node
The Load Checkpoint/Model node allows you to select the model to use, such as the Deliberate model downloaded from Civitai. You can also install and use other models.
Prompt Nodes
The Prompt nodes, such as the Positive Prompt node and Negative Prompt node, contain text that can be edited or renamed. The Positive Prompt node contains text such as "beautiful scenery nature glass ball landscape purple Galaxy ball", while the Negative Prompt node can be renamed and contains text that is not shown.
Sampler Node
The Sampler node connects to the positive and negative prompts, and lines represent connections between nodes. The Sampler node can be switched or detached and reconnected on the fly.
VAE Node
The VAE node connects to the model output and connects to the Save Image node. The Render button generates an image based on prompts and model, and the image is saved as a 512x512 image.
Image Generation Options
The Image Generation Options allow you to set the width and height in the VAE node, as well as the batch size that determines the number of images generated. The seed value in the K sampler node determines the base noise for image generation, and can be changed or set to a fixed value.
COMFY UI Tutorial
Increment and Decrement
The ix is an increment that increases one at a time or a decrement that decreases one at a time. Examples include the number of steps an image goes through while rendering, or the CFG scale, which is a mathematical equation between positive and negative prompts and output.
CFG Scale
The CFG scale determines how much the AI listens to the prompt. A lower CFG means the AI listens less to the prompt, and a recommended value is between 3-7, with 4 or 5 usually working well.
Sampler
The Sampler renders the image, and for Stabil Fusion 1.5 and 2.0 models, recommends using DPM Plus+ 2m and Scheduler: Caris.
Nodes
Nodes can be added by right-clicking, and many available nodes can be chosen from. Examples include the Cas Sampler, which has settings used to render an image, and the Advanced Cas Sampler for more settings.
Importing Images and Workflows
Comfy UI allows you to drag and drop an image into the interface to get the workflow, all nodes used to create the image, and all settings. This feature allows learning from others and iterating on their work.
COMFY UI Manager
The COMFY UI manager helps install custom nodes. Steps to install custom nodes include downloading the COMFY UI manager, extracting the contents into the COMFY UI folder, restarting Comfy UI, and using the manager to install missing custom nodes.
Comfy UI Tutorial
Introduction
This tutorial provides an overview of installing Comfy UI and setting up a custom node, as well as loading another user's image and workflow to see similar results.
Key Features
Comfy UI's node-based system provides a lot of control, freedom, and help. You can load another user's workflow and install required custom nodes, and many other custom nodes are available to explore.
Custom Nodes
Custom nodes are available, and it's recommended to try out different nodes and see what they do. Observe the green border when generating an image to see which node is active.
Learning Resources
Comfy UI's GitHub has many examples to learn from, including node setups such as in-painting node setup and Laura node setup.
In-painting Node Setup
The in-painting node setup loads a checkpoint or model, has positive and negative prompts, a sampler, and VA code for in-painting, and runs from start to finish.
By following these steps and understanding the features and capabilities of Comfy UI, you'll be well on your way to creating stunning AI-generated art and exploring the world of custom workflows.