How to Use

This Python script removes metadata from .safetensors files (commonly used for LoRAs, embeddings, and other Stable Diffusion / AI model files) and saves clean copies to a separate folder. It is especially useful for cleaning LoRAs before sharing them or rename them (just rename the original .safetensors file and run the script).

1. Prerequisites

• Python 3.8 or higher installed.

• The safetensors library installed.

Install the required package (run in your terminal / command prompt):

Bash

pip install safetensors torch

Note: You only need torch if it's not already installed, safetensors is the main dependency.

2. Prepare the Script

1. Unzip and save the provided script as metadata_remover_and_rename.py (it is already attached in this conversation).

2. Edit the folder paths at the top of the file.

Open the file in any text editor (Notepad, VS Code, etc.) and update these two lines:

Python

# Folder containing the .safetensors files (This is where your Loras are)
input_folder = r"XXXXX"   # ← Change this

# Folder for the cleaned files
output_folder = r"ZZZZZ"  # ← Change this

Example (Windows):

Python

input_folder = r"C:\Users\YourName\Downloads\MyLoRAs"
output_folder = r"C:\Users\YourName\Downloads\Cleaned_LoRAs"

Example (macOS / Linux):

Python

input_folder = r"/Users/YourName/Downloads/MyLoRAs"
output_folder = r"/Users/YourName/Downloads/Cleaned_LoRAs"

• Use r"..." (raw string) to avoid issues with backslashes on Windows.

• The script will create the output folder automatically if it doesn't exist.

3. How to Run the Script

Option A: Run from Terminal / Command Prompt (Recommended)

1. Open a terminal / command prompt.

2. Navigate to the folder where the script is saved:

Bash

cd path/to/your/script/folder

1. Run the script:

Bash

python metadata_remover_and_rename.py

or

Bash

python3 metadata_remover_and_rename.py

Option B: Run from an IDE (VS Code, PyCharm, etc.)

• Open the file and click the "Run" button (or press F5 in many editors).

4. What to Expect

• The script will process only .safetensors files in the input_folder.

• For each file it will:

  • Load the model weights.

  • Save a new copy in the output_folder without any metadata.

• You will see console output like:

text

Metadata removed: my_lora.safetensors
Metadata removed: another_model.safetensors
All files have been processed.

• Original files are not modified — they remain in the input folder.

5. Optional Improvements

Process subfolders recursively

This requires a small change (ask if you need help).

Add metadata (e.g., add your name or license)

Change the save line to:

Python

save_file(tensors, output_path, metadata={"author": "YourName", "license": "CC-BY-NC"})

6. Troubleshooting

ModuleNotFoundError: No module named 'safetensors'

Run pip install safetensors

Permission errors

Run terminal as Administrator or use paths you have write access to

Script does nothing

Make sure input_folder contains .safetensors files and paths are correct

Large files take time

This is normal — big LoRAs can take several seconds each

7. Safety Notes

• This script removes all metadata (author, trigger words, version info, etc.). Make sure you keep a backup of the original files if you need that information.

• Always test a cleaned LoRA in your favorite UI (Forge,Automatic1111, ComfyUI, etc.) to ensure it still works.