# ComfyUI Model Downloader

Download models from Hugging Face and Civitai into ComfyUI model folders. Use one YAML config and run on any machine (local, server, rented GPU).

## Setup

```bash
git clone <repo-url>
cd model-downloader
python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt
rm config.example.yaml
touch config.yaml
```

Copy the example config and edit it:

```bash
cp config.example.yaml config.yaml
```

Edit `config.yaml`: set `huggingface_token`, `civitai_token`, `comfyui_base_dir`, and add entries under `models` with `url` and `type` for each model. **Do not commit `config.yaml`** (it is gitignored).

## Run

```bash
python -m model_downloader
```

- `--config PATH` – config file (default: `config.yaml` in current directory).
- `--dry-run` – print what would be downloaded and where; no writes.
- `--only TYPE ...` – only process these model types (e.g. `--only loras diffusion_models`).

## Config and ComfyUI folders

Each model entry has `url` and `type`. The `type` value is the subdirectory name under `comfyui_base_dir/models/` (e.g. `type: loras` → `models/loras/`). Each `models/<type>` directory must already exist under your ComfyUI base; the tool does not create these directories and will fail if a directory is missing.

Tokens stay in your local `config.yaml` only; keep them out of git.

## Idempotency

Same URL and model type always write to the same path (`comfyui_base_dir/models/<subdir>/<filename>`). Re-running is safe: existing model files are skipped (no re-download, no overwrite).