Never mind the details for now. Let's just generate an image.

Launching

Portable Version

  • Double-click run_nvidia_gpu.bat in the folder where you installed it.

Desktop Version

  • Launch it from the desktop shortcut or the Start menu.

Manual Installation

  • Run the following commands:
    cd path\to\comfyui
venv\Scripts\activate
python main.py

After a short wait, ComfyUI opens in your browser.

  • If the screen does not open, try entering http://127.0.0.1:8188 in the browser's address bar.

Select Workflow from Template

  • Right after your very first launch, the template screen will probably be open. (If it isn't, select Templates in the left sidebar.)
  • First, select Getting Started -> Image Generation.
  • A Missing Models error will appear.
    • It means you don't have the models this workflow needs, but ignore it for now and press ✕ to close it.

Downloading the Model

This workflow is for running a model called Stable Diffusion 1.5. Compared to today's latest models its performance is modest... or honestly, it's not really usable, but the fundamental mechanism hasn't changed.

Let's learn the basics of image generation with this model, which you could call the ancestor of image generation AI.

  • 1. Download the model 
    📂ComfyUI/
  └── 📂models/
      └── 📂checkpoints/
          └── v1-5-pruned-emaonly-fp16.safetensors
  • 2. Refresh
    • Just placing the model is not enough for ComfyUI to recognize it.
    • Press the r key, or click ComfyUI Icon -> Edit -> Refresh Node Definitions, and it will be picked up.

Let's Generate

  • Press the ▷ Run button at the top of the screen to start generation.
  • If an image shows up in the Save Image node, it worked.
  • As the name suggests, the Save Image node saves the image it receives. Generated images are saved in this folder:
📂ComfyUI/
  └── 📂output/
      ├── SD1.5_00001_.png
      ...

Play around by changing the prompt (the text in the CLIP Text Encode node) or the seed number.

Common Problems

Model not found (Load Checkpoint is red)

  • Symptom
    • The Load Checkpoint node is outlined in red and an error appears.
    • Value not in list: ckpt_name: 'v1-5-pruned.ckpt' not in []
  • Cause
    • The specified model file is not in the models/checkpoints/ folder.
    • The file has been renamed.
  • Solution
    • Check that the file is in the right folder (models/checkpoints).
    • Renaming the file is fine, but if you do, select the model again.