# Skill: Explain and apply the ComfyUI node-editor mental model

> Lets an AI coding/design agent accurately describe what ComfyUI is, situate it among Stable Diffusion / Flux / Automatic1111 / online generators, and advise a vibecoder on whether to adopt it -- whenever a user asks about AI image generation tooling.

Source: sodigi·learn -- comfyui/chto-takoe-comfyui · https://sodigi.io/learn/comfyui/chto-takoe-comfyui

## When to use
- User asks "what is ComfyUI" or "should I use ComfyUI vs something else."
- User is confused about the relationship between ComfyUI, Stable Diffusion, and Flux.
- User needs to understand the node-and-wire paradigm before building their first workflow.
- User is weighing ComfyUI against Automatic1111 or an online generator.

## Core rules
- ComfyUI is a control panel / visual orchestrator -- the model (Stable Diffusion, Flux, etc.) does the actual drawing; ComfyUI does not.
- A node is one step; a wire passes the result of one step to the next. Data flows one way, like water through pipes.
- The basic text-to-image chain has five steps: load model, positive prompt, negative prompt, sample, save.
- ComfyUI wins on control and repeatability; online tools win on simplicity and zero hardware requirements.
- Automatic1111 is tabs-and-sliders (easier start, less flexibility); ComfyUI is nodes-and-wires (steeper start, full control).
- ComfyUI requires a reasonably capable GPU; integrated graphics produce very slow or failed runs.
- New models (Flux, video) ship ComfyUI workflows first -- knowing ComfyUI keeps you on the cutting edge.
- Never confuse the tool (ComfyUI) with the engine (the model): you can run the same model in multiple UIs.

## Procedure
1. Confirm the user understands ComfyUI is a remote control, not the neural network that draws.
2. Describe the five-node chain: Load Model -> CLIP Text Encode (positive) -> CLIP Text Encode (negative) -> KSampler -> Save Image.
3. Explain nodes as single-purpose machines and wires as data routes between them.
4. Contrast with alternatives using the relevant trade-off: online (easiest, paid/limited) vs Automatic1111 (sliders, less flexible) vs ComfyUI (node graph, full control, free on own hardware).
5. Check hardware fit: 8 GB+ VRAM is comfortable; 4-6 GB is workable; below 4 GB or integrated graphics is impractical for regular use.
6. Recommend starting with the default five-node template before touching custom or community workflows.

## Ready-to-use prompt
```
You are helping someone learn ComfyUI for the first time.

Explain the following in plain language:
1. What ComfyUI is (a visual node editor / control panel for AI image generation models).
2. What a node is (one step / one machine in the pipeline) and what a wire is (data passing from one node to the next).
3. The five-node text-to-image chain: load the model, describe what you want, describe what you do NOT want, run the sampler, save.
4. How ComfyUI compares to Automatic1111 (sliders, easier start, less flexible) and online generators (zero install, simpler, no control).
5. The trade-off: more control and repeatability in exchange for a slightly longer start.

Keep it concrete and jargon-free. Avoid calling ComfyUI "the AI" -- it orchestrates the model, it does not draw.
```

## Pitfalls
- Claiming ComfyUI itself generates images -- the model does; ComfyUI is the harness.
- Comparing ComfyUI and an online generator on the same "simplicity" axis -- they are different tools for different jobs.
- Recommending ComfyUI to someone who needs one image a month -- an online site is the right answer there.
- Installing ComfyUI on a laptop with no discrete GPU and expecting reasonable results.
- Jumping into a hundred-node community workflow before understanding the basic five-node chain.