ComfyUI

ComfyUI is a powerful, node-based graphical user interface for Stable Diffusion and other generative AI models. This cheatsheet covers installation, running, keyboard shortcuts, and troubleshooting specifically for Ubuntu environments.

---

🚀 Installation & Setup

To get ComfyUI running with GPU acceleration on Ubuntu using uv, follow these steps.

1. Prerequisites

Ensure you have the Nvidia drivers installed. You can check this by running:

nvidia-smi

Install Python, Git, and uv (a fast Python package installer and resolver written in Rust):

sudo apt update
sudo apt install python3-pip git -y

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

2. Clone Repository

Clone the official ComfyUI repository:

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

3. Virtual Environment (Recommended)

Create a Python virtual environment using uv (this will create a .venv directory):

uv venv

4. Install PyTorch with CUDA

Install PyTorch with CUDA support. Make sure to check compatibility with your installed CUDA version:

uv pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121

5. Install Dependencies

Install the remaining ComfyUI dependencies:

uv pip install -r requirements.txt

---

🏃 Running ComfyUI

Basic Start Command

Run ComfyUI using uv run (which automatically locates and uses the virtual environment):

uv run python main.py

(Alternatively, you can activate the environment via source .venv/bin/activate and run python main.py)

Command Line Arguments

Here are useful startup arguments:

Argument	Action
--listen	Listen on all network interfaces (allows access from other devices on local network)
--port 8188	Set the port (default is 8188)
--enable-manager	Enable the ComfyUI-Manager extension
--highvram / --lowvram / --novram	Force high/low/no VRAM mode (auto-detected by default)
--use-pytorch-cross-attention	Use PyTorch cross-attention (often faster on newer GPUs)
--preview-method auto	Enable previewing latent images while generating
--disable-smart-memory	Disable smart VRAM management (useful if encountering OOM crashes)

🔄 Auto-Start Script

Create a helper script run.sh in the ComfyUI root directory to quickly launch it:

#!/bin/bash
cd "$(dirname "$0")"
uv run python main.py --listen --preview-method auto --enable-manager

Make it executable:

chmod +x run.sh

---

⌨️ Keyboard Shortcuts (Web UI)

These shortcuts are essential for efficient node editing and workflow design.

Canvas Navigation & View

Shortcut	Action
Space + Drag / Right-Click + Drag	Pan/scroll the workspace
Scroll Wheel / Pinch Zoom	Zoom in / out
Ctrl + 0	Reset zoom to 100%
Alt + C	Collapse / Expand selected node

Node Editing & Creation

Shortcut	Action
Double Click (on canvas)	Open node search menu
Ctrl + Enter	Queue prompt (generate image)
Ctrl + S	Save current workflow to JSON
Ctrl + O	Load workflow from JSON
Ctrl + C / Ctrl + V	Copy and paste selected nodes (preserves connections inside copy)
Ctrl + Shift + V	Paste nodes with input connections intact
Alt + Drag (on a node)	Clone/duplicate the node
Delete / Backspace	Delete selected node(s)
Ctrl + G	Group selected nodes
Ctrl + M	Mute / unmute selected node (stops execution)

Connections & Wires

Shortcut	Action
Drag + Release (on empty space)	Open search menu filtered to compatible inputs/outputs
Ctrl + Drag (on a slot)	Move all connected wires to a different slot
Shift + Click (on a wire)	Add a reroute node (dot) to the wire
Right-Click (on input/output)	Convert between Widget (value) and Input (node connector)

---

📦 Model & Path Management

Shared Model Paths (with Stable Diffusion WebUI)

If you already run AUTOMATIC1111/Stable Diffusion WebUI, you don't need to duplicate models. You can share them.

1. Copy the template configuration file:

   cp extra_model_paths.yaml.example extra_model_paths.yaml

2. Open extra_model_paths.yaml and edit the base_path under a1111 to point to your WebUI installation:

   a1111:
       base_path: /path/to/stable-diffusion-webui

Default Model Directory Layout

Put your downloaded models in these directories under the ComfyUI root:

Model Type	Destination Folder
Checkpoints (Base Models)	models/checkpoints/
LoRA	models/loras/
VAE	models/vae/
ControlNet	models/controlnet/
Upscalers	models/upscale_models/
Embeddings (TI)	models/embeddings/

---

🛠️ ComfyUI Manager & Custom Nodes

ComfyUI-Manager is an essential custom node to install, update, and manage other custom nodes/models.

1. Install ComfyUI-Manager

Install the manager dependencies using uv:

uv pip install -r manager_requirements.txt

Enable the manager by adding the --enable-manager flag when running ComfyUI:

uv run python main.py --enable-manager

2. Manual Custom Node Installation

If you prefer manual Git installation:

cd custom_nodes
git clone <custom-node-repo-url>
cd <custom-node-folder>
# If they have requirements, install them using uv.
# uv automatically finds the .venv in the parent directory!
uv pip install -r requirements.txt

---

❓ Troubleshooting (Ubuntu Specific)

1. Out of VRAM (OOM) Errors

If you run out of GPU memory:

• Add --lowvram or --novram to your startup command.
• Set --preview-method none to disable real-time VRAM-intensive generation previews.

2. CUDA/PyTorch Version Mismatch

If ComfyUI cannot find your GPU (running on CPU instead): Ensure your PyTorch installation matches CUDA. Verify with:

python3 -c "import torch; print(torch.cuda.is_available())"

If it returns False, re-install PyTorch using the exact CUDA-enabled wheels command shown in the Installation section.

3. Monitoring GPU Usage

Open a terminal split next to ComfyUI to monitor VRAM and temperature:

watch -n 0.5 nvidia-smi