VIBE Local GUI (Windows) 🎨

A local Gradio Web-UI for the VIBE model, offering advanced controls like Seed, Sample Steps, and Guidance Scale adjustments.

Tested on Windows 11 with NVIDIA RTX 5090 (Blackwell architecture).

(Run your own local instance with full privacy and speed)

🖼️ Examples

👁️ Click here to see more examples

Object Interaction

Change Appearance

Text Removal

Object Interaction

Watermark Removal

Watermark Removal

✨ Features

• 🚀 Runs Locally: Fast, private, and no limits.
• 📊 Real-Time Monitor: (New in v1.1) Live display of CPU, RAM, and VRAM usage directly in the UI.
• 🎛️ Full Control: Adjust Sample Steps, Seed, Guidance Scale, and Image Guidance.
• 💾 Auto-Save: Results are automatically saved to the outputs folder with timestamps.
• 🛑 VRAM Friendly: Includes a Shutdown button to instantly close the server and free up GPU memory.
• 🛠️ Windows Optimized: Fixes common pathing and dependency issues for NVIDIA RTX 50 series.

⚙️ Installation

1. Prerequisites

• Python 3.10 or 3.11 installed (Make sure to check "Add Python to PATH" during installation).
• Git installed.
• An NVIDIA GPU (16GB+ VRAM recommended, 24GB+ for best performance).

2. Clone the Repository

Open PowerShell or Terminal and run:

git clone https://github.com/Detoxfox4234/VIBE-Local-GUI-Windows.git
cd VIBE-Local-GUI-Windows

3. Create a Virtual Environment

It is highly recommended to use a virtual environment to avoid conflicts.

python -m venv venv
.\venv\Scripts\activate

4. Install Dependencies

This will install gradio, torch, and the VIBE model requirements.

pip install -r requirements.txt

⚠️ Important for RTX 5090 Users (Blackwell)

The standard PyTorch version installed by requirements.txt might not yet support the RTX 50 series (Compute Capability sm_120).

If you encounter a Torch not compiled with CUDA enabled or sm_120 error, please reinstall PyTorch with the Nightly Build manually:

# 1. Uninstall the standard version
pip uninstall torch torchvision torchaudio -y

# 2. Install the Nightly version (supports CUDA 12.8+)
pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

🚀 Usage

1. Make sure your environment is active ((venv) should be visible).
2. Run the application:

python app.py

3. Wait for the model to load. Your default web browser will open automatically at http://127.0.0.1:7860.

Note: On the very first run, the script will download the VIBE model weights (~10GB) from Hugging Face. This might take a while depending on your internet connection.

📂 Output

All generated images are saved in the outputs folder inside the installation directory. The filenames include the timestamp and the seed used, so you can easily reproduce results.

🔗 Credits

• Original VIBE Model: ai-forever/VIBE
• Model Weights: Hugging Face (iitolstykh)

🤝 Support

This is a free open-source project. I don't ask for donations. However, if you want to say "Thanks", check out my profile on Spotify. A follow or a listen is the best way to support me! 🎧