# Skill: Select, place, and manage ComfyUI checkpoint models

> Lets an AI agent guide a user to pick the right checkpoint for a task, place the file correctly so ComfyUI detects it, and avoid the most common model-related failures -- whenever checkpoint loading, style mismatch, or VRAM errors are the problem.

Source: sodigi·learn -- comfyui/chekpoynty-i-modeli · https://sodigi.io/learn/comfyui/chekpoynty-i-modeli

## When to use
- User's Load Checkpoint node shows an empty list.
- Generated images do not match the expected style (e.g., anime output when a photo was expected).
- User hits an out-of-memory or crash error during generation.
- User asks where to download models or which model to choose for a specific task.

## Core rules
- A checkpoint is the neural network file that actually generates images; without one, ComfyUI produces nothing.
- Style is set by the model, not only by the prompt -- pick the right model first, then write the prompt.
- Model file must go in `ComfyUI/models/checkpoints/`; any other location will be invisible to the program.
- After placing a file while ComfyUI is running, refresh the node's dropdown or restart the program.
- File size roughly correlates with VRAM demand: larger files require more GPU memory.
- Add-ons (LoRA, ControlNet, VAE) must come from the same model family / base as the checkpoint -- cross-family add-ons are incompatible.
- Download only from established, verified platforms; read each model's license before using images commercially.
- Two or three reliable working checkpoints are enough; hoarding dozens wastes disk space without benefit.

## Procedure
1. Decide the output style needed: photorealistic, stylized (anime/cartoon), or artistic.
2. Browse a verified catalog, read example images and the description, note the model family / base.
3. Download the checkpoint file (safetensors or ckpt format).
4. Place it in `ComfyUI/models/checkpoints/` -- not in the root, not in a subfolder of checkpoints.
5. Refresh the dropdown in the Load Checkpoint node (or restart ComfyUI).
6. Select the model in the node and run a test generation with a minimal prompt.
7. If an out-of-memory error appears: try a lighter model, reduce image resolution, or enable medvram/lowvram mode.

## Ready-to-use prompt
```
I need to set up a checkpoint model in ComfyUI. Help me:

1. Choose the right type of model for my task: [describe your desired output style here].
2. Identify where to place the downloaded file in the ComfyUI folder structure.
3. Know what to check if the model does not appear in the Load Checkpoint dropdown.
4. Understand why my style does not match the prompt (likely wrong model type).
5. Know the VRAM implications of heavy vs light models.

My GPU has approximately [X] GB of VRAM.
```

## Pitfalls
- Forgetting to select any model at all -- ComfyUI will error; it is not a bug.
- Placing the file in the ComfyUI root folder or the wrong subfolder -- the dropdown will stay empty.
- Expecting an anime-trained model to produce photorealistic output by writing "photorealistic" in the prompt.
- Not refreshing the interface after dropping a file while ComfyUI is open.
- Running a large (6+ GB) model on a GPU with 4 GB VRAM without enabling memory-saving modes.
- Downloading dozens of checkpoints "just in case" -- two or three curated models outperform a cluttered library.