Essential Nodes and Components

Introduction to ComfyUI Nodes

ComfyUI operates on a node-based workflow system where each node performs a specific function in the AI image generation process. Think of nodes as building blocks that you connect together to create a complete image generation pipeline. Understanding the essential nodes is crucial for creating your first working workflows.

Core Input Nodes

Checkpoint Loader

The Checkpoint Loader is your starting point for any ComfyUI workflow. This node loads the AI model (checkpoint) that will generate your images.

Inputs: None (you select the model from a dropdown)
Outputs: MODEL, CLIP, VAE

The Checkpoint Loader provides three essential components:

• MODEL: The core diffusion model for image generation
• CLIP: Text encoder for understanding prompts
• VAE: Encoder/decoder for converting between image and latent space

CLIP Text Encode (Prompt)

This node converts your text prompts into a format the AI model can understand. You'll typically use two of these nodes:

Inputs: CLIP, text
Outputs: CONDITIONING

• Positive prompt: Describes what you want in the image
• Negative prompt: Describes what you want to avoid

Generation Nodes

KSampler

The KSampler is the heart of image generation. It performs the actual sampling process that creates your image.

Inputs: model, positive, negative, latent_image, seed, steps, cfg, sampler_name, scheduler, denoise
Outputs: LATENT

Key parameters:

• Steps: Number of denoising steps (typically 20-50)
• CFG Scale: How closely to follow your prompt (typically 7-15)
• Seed: Random seed for reproducible results
• Sampler: Algorithm used for sampling (euler, dpm++, etc.)

Empty Latent Image

This node creates a blank canvas in latent space where your image will be generated.

Inputs: width, height, batch_size
Outputs: LATENT

Standard dimensions include 512x512, 768x768, or 1024x1024 pixels.

Output Nodes

VAE Decode

Converts the generated latent image back into a viewable image format.

Inputs: samples, vae
Outputs: IMAGE

Save Image

Saves your generated image to disk.

Inputs: images, filename_prefix
Outputs: None (saves file)

Basic Workflow Connection

A minimal ComfyUI workflow follows this pattern:

1. Checkpoint Loader → provides MODEL, CLIP, VAE
2. CLIP Text Encode (x2) → converts positive/negative prompts
3. Empty Latent Image → creates canvas
4. KSampler → generates image using all above inputs
5. VAE Decode → converts latent to image
6. Save Image → outputs final result

Node Connection Tips

• Color coding: Connections are color-coded by data type (purple for models, yellow for conditioning, etc.)
• Right-click menus: Right-click nodes for additional options
• Queue system: Use "Queue Prompt" to execute your workflow
• Parameter adjustment: Double-click nodes to modify their settings

Common Beginner Mistakes

• Forgetting to connect the VAE from Checkpoint Loader to VAE Decode
• Using incompatible image dimensions
• Setting CFG scale too high or too low
• Not using negative prompts effectively

Mastering these essential nodes provides the foundation for all ComfyUI workflows, from simple image generation to complex multi-stage processes.