ComfyUI Getting Started and Detailed Guide

ComfyUI is a powerful node-based user interface for generating images via Stable Diffusion. Compared to other interfaces, ComfyUI provides a higher degree of freedom and a modular structure, suitable for users who want to deeply customize the generation process. Although it is more complex than WebUI, it provides users with a wealth of tools and plug-ins that enables you to create personalized build workflows.

1. Install ComfyUI

(1) Download ComfyUI

• GitHub official address: ComfyUI official repository

• Windows configuration-free integration package (recommended for beginners): ComfyUI Qiuye integration package (unzip ready to use, pre-installed plug-ins)

(2) Manually install ComfyUI

1. Download ComfyUI

    git clone https://github.com/comfyanonymous/ComfyUI.git
     cd ComfyUI

2. Installation dependencies

    pip install -r requirements.txt

   (Make sure Python 3.10+ and Git are installed)

3. Download Stable Diffusion Model

• SD1.5 : Hugging Face

• SDXL : SDXL model

• ControlNet (optional) : ControlNet model

• Put the .safetensors or .ckpt model files into the ComfyUI/models/checkpoints/ directory.

• Recommended model: MajicMix

4. Run ComfyUI

    python main.py

   (Windows users can directly run run_nvidia_gpu.bat

2. Recommended plug-ins

Installing plug-ins can greatly enhance the functionality of ComfyUI. The following are commonly used plug-ins:

3. ComfyUI basic operations

(1) Interface Basics

• Scroller: Zoom canvas

• Drag the blank area with the left button: Move the work area

• Double-click the blank area: Open the node search bar

• Ctrl + Drag: Select multiple nodes

• Shift + Drag: Move multiple nodes together

• Drag the input/output port: Create a connection cable (only connect to the same type of port)

(2) Run the workflow

• Click "Execution Queue" to run the current process

• After running, you can view the task progress in the "Quote" in the menu bar on the left

• After the task is completed, the generated image will be displayed below the interface. Right-click on the image to delete or load the workflow

(3) Loading workflow

• Method 1: Click "Workflow" at the top menu → "Open" and select the .json or .png file

• Method 2: Drag and drop .json or .png files directly to the ComfyUI area

4. Text-generated image workflow

ComfyUI's default SD1.5 literary graph workflow consists of multiple nodes, and the basic process is as follows:

(1) Loading model example Select the model under the ComfyUI/models/checkpoints directory

(Sample model: MajicMix realistic barley orange realistic.safetensors )

(2) Enter text prompt
CLIPTextEncode
Enter a positive prompt word (positive condition):
best quality, masterpiece, ultra high resolution, 1girl, MMN, MN, 1girl, realistic, vision, landscape, half body, close shot

Enter a negative prompt word (negative condition):
worst quality:2, low quality:2, normal quality:2, lowres, watermark, nude, mutant hands, missing arms, missing fingers, extra legs, fused fingers, too many fingers, bad hands, missing fingers, extra arms and legs

Output: Text encoding

(3) Generate latent space images
EmptyLatentImage
Set the image size (such as 512x768)
Note: According to the loading model settings, different effects are not consistent output: noise latent diagram

(4) Noise reduction sampling
KSampler
Select the sampling method (Euler a/DPM++ 2M Karras)
Setting Steps: 20
Set CFG Scale (prompt word influence): 20
Output: The latent space diagram after noise reduction

(5) Decode the image
VAEDecode
Function: Convert latent space map to visible image output: final image

(6) Save the picture
SaveImage
Save the image to the output/ directory and finally generate the image

5. Graphics (graphics with graphics) workflow

Method 1: Download the picture and the workflow

• Drag and drop workflow images to ComfyUI interface ( workflow diagram download )

• Or click the "Load" button and select the .json workflow file

Method 2: Create a graph manually

1. Load the default literary picture workflow

2. Delete an empty Latent node

3. Add the "Load Image" node (right-click → "New Node" → "Image" → "Load Image")

4. Add the "Image Scale by Coefficient" node (right-click → "New Node" → "Image" → "Zoom in")

5. Add the "VAE Encoding" node (right-click → "New Node" → "Latent" → "VAE Encoding")

6. Connect the VAE encoder to the K sampler

How to use

• Select the model, upload the picture, modify the prompt words, and adjust the sampling parameters

• Click "Execution Queue" to start generation

6. Advanced gameplay

(1) Use LoRA enhancement style to add the LoraLoader node, load the LoRA model to connect CheckpointLoaderSimple → LoraLoader → CLIPTextEncode

LoRA Download: CivitAI

(2) Use ControlNet to control the image to connect ControlNetPreProcessor → ControlNetLoader → KSampler
Enter line drawing or depth map to control the final composition

ControlNet pre-trained model: download address