ComfyUI Beginner Guide 2026: Install, Setup & First Image

ComfyUI is the most powerful free tool for running AI image generation locally — but its node-based interface confuses nearly every beginner on first launch.

This guide skips the intimidation and gets you to your first generated image step by step, whether you're on Windows or Mac.

What Is ComfyUI?

ComfyUI is a free, open-source interface for running AI image generation models on your own computer — no subscriptions, no cloud API costs, no limits on how many images you create.

Unlike Midjourney (cloud-only, subscription) or DALL-E (API costs per image), ComfyUI runs entirely on your hardware. The images it generates stay on your machine and cost you nothing after setup.

It's built around a node-based workflow system. Instead of a single text box, you connect blocks together — a model loader, a text prompt encoder, a sampler, a decoder — and the chain of nodes produces your image. This sounds complex, but it means you can build any workflow imaginable, from simple text-to-image to video generation, inpainting, and LoRA-based style transfers.

ComfyUI supports:

Stable Diffusion (all versions: SD 1.5, SDXL, SD3)
Flux (including Flux.1 Dev and Flux.1 Schnell)
Wan 2.1 (video generation)
LTX Video
Hundreds of community model variants via HuggingFace and Civitai

System Requirements

Windows:

NVIDIA GPU with at least 4 GB VRAM (8 GB recommended for Flux)
16 GB RAM minimum
10–20 GB free disk space (for models)
Can run on CPU, but extremely slow — a single image takes 10+ minutes

Mac:

macOS 12.3 or higher (for Apple Silicon MPS acceleration)
Apple Silicon (M1, M2, M3, M4) recommended — Intel Macs will be very slow
16 GB unified memory minimum; 32 GB for Flux models
Python 3.10 or later

Need to check your GPU's VRAM before starting? See our VRAM guide for a full breakdown by model size.

Installation: Windows (Desktop App — Easiest Method)

The ComfyUI Desktop app launched in 2025 and is the cleanest way to get started on Windows. It installs like a normal application — no Python, no command line required.

Step 1: Go to comfy.org/download and download the Windows installer.

Step 2: Run the installer. It sets up Python, dependencies, and the application automatically.

Step 3: Launch ComfyUI Desktop. On first run, click Get Started.

Step 4: Select your GPU type. Choose Nvidia GPU for automatic CUDA configuration, or CPU if you have no discrete GPU.

Step 5: Choose an install location — remember this path, you'll add model files here later.

Step 6: Click Install. The app downloads remaining components (~2–3 GB).

Once installed, the app opens to the main ComfyUI canvas with a default text-to-image workflow already loaded.

Installation: Windows (Portable Zip — Alternative)

If you prefer not to use the Desktop app, or you want finer control over the installation:

Step 1: Download the standalone portable version from the ComfyUI GitHub releases.

Step 2: Install 7-Zip if you don't have it. Right-click the downloaded .7z file → Extract Here.

Step 3: Download a model checkpoint. For beginners, Dreamshaper 8 is a good starting point. Place the downloaded .safetensors file in:

ComfyUI_windows_portable\ComfyUI\models\checkpoints

Step 4: Launch ComfyUI by double-clicking run_nvidia_gpu.bat (or run_cpu.bat if no GPU). This opens a browser tab at localhost:8188.

Installation: Mac

Mac installation requires using the Terminal. If you haven't used Terminal before, see our Terminal beginner guide first — it covers everything you need.

Step 1: Install Homebrew if you don't have it:

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

Step 2: Install required tools:

brew install cmake protobuf rust python@3.10 git wget

Step 3: Clone ComfyUI from GitHub:

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

Step 4: Create a virtual environment and install dependencies:

python3 -m venv venv
./venv/bin/pip install torch torchvision torchaudio
./venv/bin/pip install -r requirements.txt

Step 5: Download a model checkpoint and place it in the models/checkpoints folder inside your ComfyUI directory.

Step 6: Start ComfyUI:

./venv/bin/python main.py

ComfyUI opens at localhost:8188 in your browser.

Understanding the Node System

When ComfyUI first loads, the canvas shows a default workflow with several connected boxes. Here's what they do:

Load Checkpoint — loads your AI model file. This is where you select which model you want to use for generation.

CLIP Text Encode (x2) — converts your text prompt (and negative prompt) into numbers the model understands. One node is for what you want in the image; one is for what you want excluded.

KSampler — the core generation node. It runs the diffusion process, turning random noise into your image over multiple steps. Key settings here: steps (more = higher quality, slower), cfg (how strictly the model follows your prompt), and sampler_name (the algorithm used).

VAE Decode — converts the internal representation of the image into actual pixel data you can see.

Save Image — saves the generated image to your disk.

How to run: Click the Queue Prompt button at the top right. ComfyUI processes the nodes in sequence and displays the output image on the right side of the canvas.

Installing ComfyUI Manager

ComfyUI Manager is the essential extension that lets you install custom nodes (community-built add-ons) with one click. Without it, you'd need to manually clone Git repositories.

Install ComfyUI Manager:

Open Terminal (Mac) or Command Prompt (Windows), navigate to your ComfyUI folder
Run:

cd custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager

Restart ComfyUI

After restart, a Manager button appears in the top-right corner. From here you can:

Browse and install community node packs
Update all installed nodes at once
Enable/disable nodes without deleting them
Install missing nodes detected in shared workflows

Running Flux Models (2026 Standard)

Flux.1 is the 2026 standard for high-quality image generation — sharper details, better text rendering, and more accurate anatomy than earlier Stable Diffusion models.

What you need for Flux.1 Schnell (faster, open license):

flux1-schnell.safetensors checkpoint
clip_l.safetensors and t5xxl_fp8_e4m3fn.safetensors CLIP models
ae.safetensors VAE
Minimum 8 GB VRAM (12 GB recommended)

Download model files from HuggingFace and place them in the corresponding models/ subfolders.

For Flux, use a Flux-specific workflow template — the default SD workflow won't work. The ComfyUI Desktop app includes Flux workflow templates in the Templates Gallery.

ComfyUI vs Automatic1111: Which Should You Use?

Feature	ComfyUI	Automatic1111
Interface	Node-based (visual)	Traditional form UI
Learning curve	Steeper initially	Easier to start
Workflow flexibility	Unlimited	High but fixed
VRAM efficiency	Excellent	Good
Flux support	Native, excellent	Limited
Community extensions	Largest ecosystem	Large ecosystem
Best for	Power users, advanced workflows	Beginners wanting simpler UI

If you're just starting with AI image generation, Automatic1111 has a gentler learning curve. If you want the most capability and plan to use newer models like Flux, ComfyUI is where to start.

Common First-Run Errors

"No models found" — you haven't placed a model file in the models/checkpoints folder. Download a .safetensors file and add it there, then restart ComfyUI.

"CUDA out of memory" — your GPU doesn't have enough VRAM for the model. Try adding --lowvram to your launch command, or use a smaller model.

Blank canvas on Mac — Apple Silicon requires the --force-fp16 flag. Add it when launching: ./venv/bin/python main.py --force-fp16

Nodes appear red/disconnected — you're loading a workflow that requires custom nodes you haven't installed. Use ComfyUI Manager → Install Missing Nodes to fix automatically.

FAQ

Q: Is ComfyUI free?
A: Yes, completely free and open source. The models you download are also free (with varying licenses). Flux.1 Schnell is Apache 2.0 (fully open). Flux.1 Dev requires a non-commercial license.

Q: Can I run ComfyUI without a GPU?
A: Yes, on CPU only — but expect 10–30 minutes per image rather than seconds. For regular use, a dedicated GPU (NVIDIA on Windows/Linux, Apple Silicon on Mac) is effectively required.

Q: Does ComfyUI work on Windows 11?
A: Yes, fully supported. The Desktop app is the easiest option on Windows 11.

Q: What's the difference between ComfyUI Desktop and the portable version?
A: Desktop is a proper installed application with a GUI setup wizard. The portable version is unzipped and run via a batch file. Desktop is easier; portable gives more control over file locations.

Q: What's ComfyUI Manager?
A: An extension that adds a package manager to ComfyUI. It lets you install, update, and remove community-built nodes without using the command line.

Q: What are the best models for beginners in 2026?
A: Dreamshaper 8 (photorealistic, forgiving), Juggernaut XL (SDXL, portrait quality), or Flux.1 Schnell (2026 quality standard, requires 8 GB VRAM).

Q: Can I use ComfyUI on a laptop?
A: Yes — if your laptop has a dedicated NVIDIA GPU with 4+ GB VRAM or is an Apple Silicon Mac. Gaming laptops with RTX 3060 or higher run ComfyUI well.