Introduction to ComfyUI
What is ComfyUI?
Section titled “What is ComfyUI?”ComfyUI is a node-based AI image/video generation workflow tool. You can visually build complex AI generation pipelines by connecting nodes, each responsible for a specific function.
Core Concepts
Section titled “Core Concepts”Node
- The basic unit of a workflow. Each node performs a single function.
- The left side of a node is the input, and the right side is the output.
For detailed explanations, please refer to the “Understanding Parameters” section below.
Workflow
Here is what a basic workflow looks like. A workflow is the entire task flow created by connecting nodes. It can be saved and shared as a JSON file.
Subgraph A subgraph is a module that groups multiple nodes together. It helps organize complex workflows neatly and allows for reuse.
Basic Workflow Structure
Section titled “Basic Workflow Structure”All image generation workflows follow this flow:
Main Node Types
Section titled “Main Node Types”| Node | Role | Description |
|---|---|---|
| Load Checkpoint | Load Model | Loads the model to be used |
| CLIPTextEncode | Prompt Encoding | Enters the text prompt |
| EmptyLatentImage | Create Empty Canvas | Sets the image dimensions |
| KSampler | Sampling (Core) | Configures detailed settings for image generation |
| VAEDecode | Decoding | Generates the actual image from configured values |
| SaveImage | Save | Saves the generated image to a file |
Main Data Types
Section titled “Main Data Types”The following data types flow between nodes:
| Type | Description | Color (ComfyUI Default) |
|---|---|---|
| MODEL | Diffusion model (UNET) | Purple |
| CLIP | Text encoder model | Yellow |
| VAE | Image encoder/decoder | Red |
| CONDITIONING | Encoded prompt/conditions | Orange |
| LATENT | Latent space image data | Pink |
| IMAGE | Actual pixel image (RGB) | Blue |
| MASK | Mask (for inpainting) | Green |
Understanding Key Parameters
Section titled “Understanding Key Parameters”KSampler Parameters
Section titled “KSampler Parameters”These are the most important parameters that determine the quality and outcome of image generation:
| Parameter | Role | Typical Values |
|---|---|---|
| seed | Random seed. Same value = same result | Any number |
| steps | Number of generation steps. Higher = more refined but slower | 4-30 (varies by model) |
| cfg | Prompt adherence strength | 1.0-11.0 (varies by model) |
| sampler_name | Sampling algorithm | euler, dpmpp_2m, etc. |
| scheduler | Noise scheduler | normal, simple, sgm_uniform, etc. |
| denoise | Denoise strength (important for img2img) | 0.0-1.0 |
Image Generation Model History (Image Model Landscape)
Section titled “Image Generation Model History (Image Model Landscape)”| Generation | Model | Released | Resolution | Minimum VRAM |
|---|---|---|---|---|
| 1st Gen | SD 1.5 | 2022.10 | 512x512 | 4GB |
| 2nd Gen | SDXL | 2023.07 | 1024x1024 | 8GB |
| 3rd Gen | SD 3.5 | 2024 | 1024x1024 | 8GB |
| 4th Gen | Flux | 2024.08 | 1024x1024+ | 12GB |
| 5th Gen | Qwen Image | 2025 | Various | 12GB+ |
| 5th Gen | Z-Image | 2025 | 1024x1024 | 8GB |
Guidebook Structure
Section titled “Guidebook Structure”This guidebook is organized by task type. Refer to the relevant chapter based on the task you want to accomplish.
| Chapter | Content | Difficulty |
|---|---|---|
| 01. Image Generation | Text-to-Image | Beginner |
| 02. Image Editing | Inpainting, Outpainting, Relighting, Background Removal | Elementary |
| 03. Guided Generation | ControlNet, LoRA, Reference-based Generation | Intermediate |
| 04. Upscale/Enhancement | Resolution Enhancement | Elementary |
| 05. Video Generation | Text-to-Video, Image-to-Video | Intermediate |