ComfyUI LoRA Metadata Scraper

This is a Python tool to recursively extract metadata from .safetensors files inside your ComfyUI installation. It supports both local metadata extraction and optional CivitAI scraping (for trigger words, tags, descriptions, and preview images/videos).

Features:

Extracts metadata from LoRAs, checkpoints, VAEs, and ControlNets
Optionally scrapes CivitAI data (tags, trigger words, preview images/videos)
Skips files that are already processed (unless forced)
Lets you scan only the LoRAs folder or the entire ComfyUI setup
Saves previews in subdirectories or alongside model files
Supports downloading all, none, or a specific number of previews
Fully interactive OR command-line friendly
Gracefully handles CivitAI rate limiting and retries

Folder Setup:

Place everything inside:

ComfyUI/
└── utils/
    └── Scrapertool/
        ├── scrape_metadata_recursive.py
        ├── requirements.txt
        └── run_scraper.bat  (optional launcher)

Usage:

Option 1: Interactive Mode

Run:

python scrape_metadata_recursive.py --interactive

You’ll be prompted:

A) Scrape CivitAI? (Y/N) B) Use default delay (0.5s), no delay (0), or custom delay? C) Force re-scrape if metadata already exists? (Y/N) D) Scan only the LoRA folder? (Y/N) E) Save previews in a subdirectory? (Y/N) F) How many preview images/videos to download? (A=All, N=None, or a number)

Option 2: Command-Line Arguments

Example:

python scrape_metadata_recursive.py --scrape-civitai --delay 0.5 --force --loras-only --previews-subdir --max-media 5

Command-Line Arguments:

--interactive Launch interactive menu.

--scrape-civitai Enable scraping of metadata and previews from CivitAI.

--delay [seconds] Set the delay between API calls and image/video downloads (default: 0.5 seconds).

--force Force re-scrape even if metadata JSON already exists.

--loras-only Scan only the ComfyUI/models/loras folder. If omitted, scans all subdirectories.

--previews-subdir Save previews inside a dedicated subdirectory (e.g., model_previews/).

--no-previews-subdir Save previews directly alongside the model file.

--max-media [number] Set max number of preview images/videos to download (0 = none, default = all).

How It Finds CivitAI Metadata:

First, the script checks if the .safetensors file contains a civitai_metadata block (fixed in the latest version to look in the correct place).
If that block is missing, it falls back to computing the SHA256 hash and queries CivitAI using that.

This ensures compatibility even if metadata is incomplete.

Dependencies:

safetensors torch requests tqdm

Setup Instructions:

Create a virtual environment:

python -m venv scrapervenv

Activate the virtual environment:

On Windows: scrapervenv\Scripts\activate

On macOS/Linux: source scrapervenv/bin/activate

Install dependencies:

pip install -r requirements.txt

requirements.txt:

safetensors torch requests tqdm

License:

Creative Commons Attribution 4.0 International (CC BY 4.0)

You may use, share, and adapt this tool for any purpose—including commercial use—as long as you provide attribution to the original creator.

Contributing:

Contributions and improvements are welcome. Please fork, submit pull requests, or open an issue if you'd like to suggest features or report bugs.