Leaving the details aside, let's generate an image for now.

Startup

Portable Version

  • Double-click run_nvidia_gpu.bat in the installed folder.

Desktop Version

  • Start from the desktop shortcut or Start menu.

Manual Installation

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

After a while, the ComfyUI screen will open 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

  • Immediately after the very first startup, the template screen is probably open. (If not, select Templates on the left sidebar.)
  • First, select Getting Started -> Image Generation.
  • An error Missing Models will be displayed.
    • This indicates that the model for running the workflow is missing, 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 current latest models, its performance is modest... or rather, honestly, it's not very useful, but the fundamental mechanism hasn't changed.

Let's learn basic image generation with this model, which can be said to be the ancestor of image generation AI.

    1. Download Model
    📂ComfyUI/
  └── 📂models/
      └── 📂checkpoints/
          └── v1-5-pruned-emaonly-fp16.safetensors
    1. Refresh
    • Just placing the model is not enough for ComfyUI to recognize it.
    • Press the r key on your keyboard, or press ComfyUI Icon -> Edit -> Refresh Node Definitions to recognize it.

Let's Generate

  • Press the ▷ Run button at the top of the screen to start generation.
  • If an image is displayed in the Save Image node, it is successful.
  • The Save Image node is, as the name suggests, a node that saves the input image. The generated image is saved in the following folder:
📂ComfyUI/
  └── 📂output/
      ├── SD1.5_00001_.png
      ...

Try changing the prompt (text in the CLIP Text Encode node) or changing the seed number to play around.

Common Troubles

Model not found (Load Checkpoint is red)

  • Symptom
    • The Load Checkpoint node is surrounded by red and an error is displayed.
    • 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 name has been changed.
  • Solution
    • Check if the folder location is correct (models/checkpoints).
    • There is no problem if the file name changes, but in that case, please re-select the model.