10 Essential ComfyUI Nodes Every Beginner Should Know

What are Load nodes in ComfyUI?

Load nodes are how you bring anything from outside into your ComfyUI workflow. Images, videos, AI models, LoRAs, VAEs, text encoders. If it needs to enter the workflow, it goes through a Load node.

Double-click your canvas and type "load" to see all available options. The most common ones:

Load Image

A core ComfyUI node. Upload a file or drag and drop an image straight onto the canvas. The output is a blue image connection you can route anywhere that accepts images.

Load Video (VHS)

ComfyUI has a built-in Load Video node, but most people use the Video Helper Suite (VHS) version instead. The difference is significant. The VHS version gives you individual frame output, frame count, audio, and full video metadata. The core version gives you one video out.

The VHS Load Video node also lets you control frame rate, resize dimensions, skip frames from the start, select every Nth frame, and limit total frames loaded. If you're doing AI video work, this is the node you want.

Load Diffusion Model / Load LoRA / Load CLIP / Load VAE

Each model type has its own Load node. The color of the output tells you what's being sent: purple for models, yellow for CLIP (text encoders), red for VAE. A LoRA node has both a model strength and a CLIP strength, letting you control how much it affects generation versus text interpretation.

What does the KSampler node do in ComfyUI?

The KSampler is the engine of your AI workflow. It takes a model, a positive prompt, a negative prompt, and a latent image, then runs the generation process. This is where the AI image or video gets created.

There are multiple KSampler variants (Advanced, Efficient, custom versions), but the standard KSampler covers most use cases. Here's what each setting controls:

Seed: The random starting noise for generation. Set to "randomize" for different results each time. Set to "fixed" with a specific number to reproduce the same output. Same seed + same settings = same image.

Steps: How many denoising passes the generation runs. Think of it like frames in an animation from pure noise to final image. More steps means more refinement, but slower generation.

CFG (Classifier Free Guidance): Controls how closely the generation follows your prompt. Setting it to 1 disables the negative prompt and roughly doubles generation speed. Most models have a recommended CFG value.

Sampler name and Scheduler: Different algorithms for the denoising process. Think of sampler as which type of hammer you use, and scheduler as whether you swing hard at the start or end. For most workflows, the default values work fine.

Denoise: This is the one setting you need to understand deeply. It's a value between 0 and 1. At 1.0, the AI completely rebuilds whatever input it receives. At 0, the input passes through unchanged. At 0.5, the AI changes roughly half the information. This is how you control image-to-image strength in any workflow.

How do Save Image and Preview Image nodes work in ComfyUI?

Save Image writes your generated image to disk. Preview Image shows you the result without saving. Both accept any blue image connection as input.

In the Save Image node, you set a filename prefix. Add forward slashes to create folder structures. For example, typing project-name/renders/image saves to nested folders automatically.

Use Preview Image when you're testing and iterating. It shows the same image at the same resolution. It does everything Save Image does except write to disk. This keeps your output folder clean during experimentation.

What do VAE Encode and VAE Decode do in ComfyUI?

VAE Encode converts a visible image into latent space (where the AI works). VAE Decode converts it back into a visible image. They're the translators between the human world and the AI world.

Think of it this way. You have two universes: the RGB universe (images you can see) and the latent space (where AI models operate). The VAE Encode sends an image into the latent space so AI models can process it. The VAE Decode brings it back so you can see the result.

A typical flow looks like this: Load Image → VAE Encode → KSampler → VAE Decode → Save Image. The VAE itself is loaded through a Load VAE node and connected to both the encode and decode steps.

What is the CLIP Text Encode node in ComfyUI?

CLIP Text Encode is your prompt box. You type a text description, and the node converts it into a conditioning signal the AI model can understand. It takes a CLIP model input and outputs a conditioning connection.

You'll typically use two of these: one for the positive prompt (what you want) and one for the negative prompt (what you want to avoid). Both connect to the KSampler.

You'll see workflows where these nodes are colored green and red. Green for positive, red for negative. But the color is cosmetic. The workflow creator chose those colors as a visual aid. You can connect either color to either input. What matters is which KSampler input it's wired to, not the node color.

What is the Empty Latent Image node in ComfyUI?

Empty Latent Image creates a blank canvas in latent space at whatever dimensions you specify. Use it when you're generating an image from scratch (text-to-image) instead of working from an existing image.

Set your width and height in pixels. The output connects to the KSampler's latent input. This tells the KSampler the dimensions of the image you want to generate. You can set it to any size: 512x512 for a quick square, 1024x768 for landscape, whatever your workflow needs.

If you're loading an image as input (image-to-image), you don't need this node. The loaded image provides the canvas. If you're starting from text alone, this node is required.

How do you resize images in ComfyUI?

The Resize Image node (the KJNodes "Resize Image V2" is a popular choice) lets you change image dimensions with control over cropping, aspect ratio, and proportion handling. Most workflows need images at specific sizes, so this node shows up constantly.

Set your target width and height. Then choose how to handle the aspect ratio: crop to fit, resize proportionally, stretch (usually not what you want), or pad the edges with a border. When cropping, you can set the crop position (center, top, bottom) to keep the important part of the image visible.

The node also outputs width and height values as separate data. Useful when other nodes in your workflow need to know the image dimensions. Some workflows also require dimensions divisible by a certain number (like 8 or 64), and this node has a setting for that.

What is the Primitive node in ComfyUI?

The Primitive node is a universal control. Connect it to any input on any node, and it adapts to that input type. Change the value in one place, and every connected node updates. It's how you build workflows with centralized controls.

Say you have three KSamplers in a workflow and want them all to use the same number of steps. Connect a Primitive node to the "steps" input on each one. Now you change the step count in one place instead of three.

It works with any input type. Drag it onto a sampler name input, and it becomes a sampler selector. Drag it onto a number field, and it becomes a number input. You can even set it to auto-increment after each generation. This makes it useful for batch testing with systematic changes.

What are the Note and Preview Any nodes in ComfyUI?

The Note node is a text box you can place anywhere on your canvas for instructions, reminders, or documentation. The Preview Any node displays text output from any node in your workflow, so you can see what's happening inside the pipeline.

Note

Place it next to complex sections of your workflow with instructions. When you share workflows with colleagues or the community, notes help others understand what each section does without guessing. Also useful as reminders for your future self.

Preview Any

This node accepts any text or string output and displays it. Useful for debugging. If you have an LLM node generating prompts, connect its output to Preview Any to see what it produced before sending it into the rest of your workflow. Any node that outputs text or metadata can be previewed this way.

What are Set and Get nodes in ComfyUI?

Set and Get nodes let you send data across your workflow without drawing connection wires. The Set node saves a value under a name you choose. The Get node retrieves it anywhere else on the canvas. They keep complex workflows clean and readable.

Send an image into a Set node and label it "source_image." Now anywhere in your workflow, drop a Get node, select "source_image," and that image appears. No wire needed. One Set node can feed multiple Get nodes, so a single input can go to five different places without five overlapping wires.

You'll see these in most community workflows. They're the reason some shared workflows look clean instead of like a bowl of spaghetti.

How do these nodes work together in a real ComfyUI workflow?

A typical text-to-image workflow connects these nodes in a straight line: load models, set up prompts, define canvas size, run the KSampler, decode the result, and save the image.

Here's the flow:

1. Load Diffusion Model loads your AI model (e.g., Stable Diffusion, SDXL, Flux)
2. Load CLIP loads the text encoder for your model
3. Load VAE loads the image encoder/decoder
4. Set nodes store model, clip, and VAE as named variables
5. Get nodes pull CLIP into two CLIP Text Encode nodes (positive and negative prompts)
6. Empty Latent Image defines your output dimensions
7. Everything feeds into the KSampler, which runs the generation
8. VAE Decode converts the output from latent space to a visible image
9. Save Image writes the result to your output folder

Every complex workflow is a variation on this pattern. Video generation, ControlNet, IP-Adapter, style transfer. They all add nodes between these core steps. Understanding this foundation makes it possible to read any workflow you download.

Frequently Asked Questions