# Skill: Install ComfyUI and configure it for first use

> Lets an AI agent guide a user through picking an install method, verifying GPU fit, placing files in the correct folders, and getting ComfyUI Manager in place -- whenever someone is setting up ComfyUI from scratch.

Source: sodigi·learn -- comfyui/ustanovka-comfyui · https://sodigi.io/learn/comfyui/ustanovka-comfyui

## When to use
- User is installing ComfyUI for the first time and does not know which method to choose.
- User gets an "out of memory" error or a blank model list after install.
- User cannot find a downloaded model inside ComfyUI's interface.
- User asks where to put model files or custom nodes.

## Core rules
- Three install paths: desktop app (easiest, stable), portable build (unzip and run, Windows-focused, fresher versions), manual install (terminal required, most control, not for day one).
- Always recommend the desktop app or portable build to beginners; save manual for later.
- VRAM is the bottleneck -- it is GPU memory, not system RAM; 8 GB+ is comfortable, 4-6 GB workable, under 4 GB very limited.
- Models go in `ComfyUI/models/checkpoints/`; LoRAs in `models/loras/`; ControlNet in `models/controlnet/`; custom nodes in `custom_nodes/`.
- Never place a model file in the root or in an arbitrary folder -- ComfyUI will not see it.
- Install ComfyUI Manager immediately after the base install; without it, community workflows will fail with missing red nodes.
- After dropping a file into a folder while ComfyUI is running, refresh the node's list or restart the program.

## Procedure
1. Ask or determine the user's OS and GPU VRAM to pick the right install path.
2. For beginners: download the official desktop installer, run it, accept defaults.
3. Open ComfyUI -- the default workflow should appear automatically.
4. Install ComfyUI Manager (via the built-in interface button or the Manager's own install page) and restart.
5. Download one checkpoint model from a trusted source and place it in `models/checkpoints/`.
6. In the Load Checkpoint node, select the file from the dropdown (refresh if it does not appear).
7. Click Queue (run). If an out-of-memory error appears, reduce image size and retry; or relaunch with medvram/lowvram flags.

## Ready-to-use prompt
```
I am setting up ComfyUI for the first time. Walk me through the process step by step:

1. Which install method to pick (desktop app vs portable vs manual) and why.
2. What VRAM I need and what to do if my GPU is weak.
3. How to install ComfyUI Manager right after the base install.
4. Exactly where to place a downloaded checkpoint model so ComfyUI can see it.
5. What to do if the model does not appear in the dropdown list.
6. How to run my first image and what to do if I hit an out-of-memory error.

My setup: [describe your OS and GPU here]
```

## Pitfalls
- Starting with the manual install (terminal + Python) before trying the desktop app -- too many failure points for day one.
- Ignoring VRAM requirements and downloading the largest available model on a 4 GB card.
- Confusing system RAM with GPU VRAM -- only GPU VRAM matters for model loading.
- Skipping ComfyUI Manager -- community workflows break without it.
- Dropping the model file into the root ComfyUI folder instead of `models/checkpoints/`.
- Not refreshing the model list after adding a file while ComfyUI is already open.