I have previously introduced Stable Diffusion webUI, which is currently the most mainstream version used by everyone: AUTOMATIC1111.

A few days ago, I also shared how to control Stable Diffusion image generation with Python.

Today, let's talk about ComfyUI: a node-based (node flow style) user interface for Stable Diffusion.

It roughly looks like this:

Let's compare AUTOMATIC1111 (WebUI) and ComfyUI.

I wrote a script and ran it.

https://github.com/comfyanonymous/ComfyUI/blob/master/notebooks/comfyui_colab.ipynb

It's still quite clear and can help understand the structure of stable diffusion. This process was shared by someone on Zhihu. (https://zhuanlan.zhihu.com/p/620297462)

The general understanding is that through individual nodes, you can control the prompt, model, view, size, etc., to generate the final image.